diff --git a/docs/public/core-concepts/models.mdx b/docs/public/core-concepts/models.mdx index 7bda6217b..c466e3f96 100644 --- a/docs/public/core-concepts/models.mdx +++ b/docs/public/core-concepts/models.mdx @@ -9,6 +9,43 @@ No single model is best at everything. Fabro lets you assign the right model to Ensemble workflow: fan out to Opus and Gemini Pro, merge, then synthesize +## How model selection works + +Fabro separates the name a workflow uses from the value a provider expects on +the wire: + +| Term | Meaning | +|---|---| +| **Provider ID** | Who serves the request, such as `openai` or `openrouter`. | +| **Model slug** | The canonical, human-facing model ID, such as `gpt-5.6-sol`. | +| **Alias** | An alternate user-facing selector, such as `gpt-56-sol`. | +| **Offering** | One provider's route to one model slug. Its identity is `(provider, model slug)`. | +| **Family** | Metadata used for display and compatible-model matching, not a routing namespace. | +| **API ID** | The opaque string sent to the selected provider API. Workflows do not reference it. | + +A model slug is unique within a provider, not across the whole catalog. Two +providers can offer the same slug and reuse the same alias, so a workflow can +use one stable selector wherever either provider is available. + +For an unqualified selector, Fabro first finds matching offerings on **ready +providers**—providers whose adapters registered successfully with usable +credentials and configuration. It then chooses the provider with the highest +`priority`; equal priorities use canonical provider ID in ascending order. A +canonical model-slug match is considered before alias matches. + +| Ready providers | Selector | Selected offering | +|---|---|---| +| OpenAI only | `gpt-56-sol` | OpenAI's `gpt-5.6-sol` | +| OpenRouter only | `gpt-56-sol` | OpenRouter's `gpt-5.6-sol` offering | +| OpenAI and OpenRouter | `gpt-56-sol` | OpenAI, because its provider priority is higher | +| Both, with `provider = "openrouter"` | `gpt-56-sol` | OpenRouter, because an explicit provider is a pin | + + +An explicit provider restricts lookup to that provider. If the pinned provider +is unavailable or does not offer the selector, Fabro reports the error instead +of silently switching providers. + + ## Model catalog | Model | Provider | Aliases | Context | Cost (in/out per Mtok) | Speed | @@ -46,13 +83,14 @@ Claude Fable 5 is available as an explicit model but is not the default Anthropi ## Configuring providers and models -Fabro's catalog starts with the built-in providers and models, then merges any `[llm]` entries from settings. Provider and model IDs are strings, so a server or project can add an OpenAI-compatible provider without a Fabro release. +Fabro's catalog starts with the built-in providers and models, then merges any `[llm]` entries from settings. Provider and model IDs are strings, so a server or project can add an OpenAI-compatible provider without a Fabro release. Declare each model under the provider that serves it: ```toml title="settings.toml" [llm.providers.proxy] display_name = "Acme Gateway" adapter = "openai_compatible" base_url = "https://llm-gateway.example.com/v1" +priority = 50 aliases = ["gateway"] [llm.providers.proxy.auth] @@ -62,8 +100,7 @@ credentials = ["env:ACME_GATEWAY_API_KEY", "vault:ACME_GATEWAY_API_KEY"] x-portkey-api-key = "{{ env.PORTKEY_API_KEY }}" x-portkey-config = "@bedrock-prod" -[llm.models."team-code-large"] -provider = "proxy" +[llm.providers.proxy.models."team-code-large"] api_id = "provider-wire-model-name" agent_profile = "anthropic" display_name = "Team Code Large" @@ -73,32 +110,33 @@ small_default = true aliases = ["team-code"] estimated_output_tps = 80 -[llm.models."team-code-large".limits] +[llm.providers.proxy.models."team-code-large".limits] context_window = 200000 max_output = 32000 -[llm.models."team-code-large".features] +[llm.providers.proxy.models."team-code-large".features] tools = true reasoning = true reasoning_effort = "levels" prompt_cache = true -effort = true -[llm.models."team-code-large".controls] +[llm.providers.proxy.models."team-code-large".controls] reasoning_effort = ["low", "medium", "high"] speed = ["fast"] -[llm.models."team-code-large".costs] +[llm.providers.proxy.models."team-code-large".costs] input_cost_per_mtok = 1.50 output_cost_per_mtok = 8.00 cache_input_cost_per_mtok = 0.30 -[llm.models."team-code-large".costs.speed.fast] +[llm.providers.proxy.models."team-code-large".costs.speed.fast] input_cost_per_mtok = 3.00 output_cost_per_mtok = 16.00 cache_input_cost_per_mtok = 0.60 ``` +The table key (`team-code-large`) is the model slug that workflows select. `api_id` is an opaque provider-facing wire value. It defaults to the exact model slug when omitted, so configure it only when the provider expects a different string, such as a deployment name, `author/model` slug, or Bedrock inference-profile ID. Fabro does not parse it for provider routing, add prefixes, or otherwise infer meaning from it; an explicitly empty `api_id` is invalid. + For [LiteLLM](/integrations/litellm), Fabro ships a disabled provider entry. Enable it in settings and declare the models your proxy exposes: ```toml title="settings.toml" @@ -106,24 +144,27 @@ For [LiteLLM](/integrations/litellm), Fabro ships a disabled provider entry. Ena enabled = true base_url = "http://localhost:4000/v1" -[llm.models."litellm-gpt-5"] -provider = "litellm" +[llm.providers.litellm.models."litellm-gpt-5"] api_id = "gpt-5" display_name = "LiteLLM GPT-5" family = "litellm" default = true -[llm.models."litellm-gpt-5".limits] +[llm.providers.litellm.models."litellm-gpt-5".limits] context_window = 128000 max_output = 8192 -[llm.models."litellm-gpt-5".features] +[llm.providers.litellm.models."litellm-gpt-5".features] tools = true vision = false reasoning = false ``` -`api_id` is the model name sent to the provider API. Omit it when the Fabro model ID and provider model ID are the same. +### Reusing aliases across providers + +Aliases are scoped to a provider. An alias or canonical slug must identify exactly one model within that provider, so two models under `proxy` cannot both claim `team-code`. The same slug or alias may be reused by another provider; that reuse is what makes unqualified selectors portable. Across providers, an exact canonical-slug match takes precedence over an alias match. You can still reach a shadowed alias by pinning its provider. + +The older `[llm.models.]` form remains readable as a compatibility input, but new configuration and built-in catalog entries should use `[llm.providers..models.]`. Model roles are separate: `default = true` controls normal model selection for workflow execution, while `small_default = true` marks the provider's small/cheap utility model for metadata tasks such as generated run titles. If a provider has no small default, Fabro falls back to that provider's normal default. @@ -169,7 +210,7 @@ Fabro ships an Ollama provider definition that is disabled by default. Enable it enabled = true ``` -Enabling the provider alone does not expose any models — until #267 adds auto-discovery, add explicit `[llm.models.]` blocks for each Ollama model you have pulled locally. Ollama's OpenAI-compatible endpoint accepts any bearer token, so local users can set `OLLAMA_API_KEY=ollama`. +Enabling the provider alone does not expose any models — until #267 adds auto-discovery, add an explicit `[llm.providers.ollama.models.]` block for each Ollama model you have pulled locally. Ollama's OpenAI-compatible endpoint accepts any bearer token, so local users can set `OLLAMA_API_KEY=ollama`. ## Default models @@ -217,11 +258,12 @@ Model stylesheets set per-node models inside the workflow graph, but you can als Pass `--model` and optionally `--provider` to `fabro run`: ```bash -fabro run docs/internal/demo/01-hello.fabro --model claude-opus-4-6 +fabro run docs/internal/demo/01-hello.fabro --model gpt-56-sol +fabro run docs/internal/demo/01-hello.fabro --model gpt-56-sol --provider openrouter fabro run docs/internal/demo/04-pipeline.fabro --model gemini-3.1-pro-preview ``` -These flags set the default model for all nodes that don't have an explicit model assigned via a stylesheet. The provider is automatically inferred from the model catalog — you only need `--provider` for models not in the catalog or to force a specific provider. +These flags set the default model for all nodes that don't have an explicit model assigned via a stylesheet. Without `--provider`, the selector is portable across ready offerings and provider priority decides. `--provider` is an explicit pin, including for models not in the catalog. ### Run config TOML @@ -247,7 +289,13 @@ Then launch with: fabro run run.toml ``` -The `fallbacks` array is optional. Each entry may be a bare provider token (like `"gemini"`), a bare model alias (like `"gpt-5.4"`), or a qualified `"provider/model"` reference. Fabro tries them in order when the primary provider is unavailable. +The `fallbacks` array is optional. Each entry may be a bare provider token (like `"gemini"`), a bare model alias (like `"gpt-5.4"`), or a qualified `"provider/model"` reference. Fabro resolves each entry to a concrete provider and canonical model, then tries that persisted chain in order after a failover-eligible error. Provider-only entries choose the closest compatible model; qualified model entries stay pinned to their named provider. + +### Resolution is stable for a run + +When Fabro creates a run, it resolves every implicit model selector against the ready-provider snapshot and persists the chosen canonical `(provider, model slug)` in the run. Resuming that run uses the materialized choice—it does not re-rank providers because credentials or priorities changed later. + +This resolve-once behavior makes a run reproducible; the fallback chain is the separate mechanism for handling a provider that fails after creation. A newly created run can choose a different ready offering from the same portable selector. The precedence order is: node-level stylesheet > run config TOML > CLI flags > server defaults. More specific settings always win. diff --git a/docs/public/execution/failures.mdx b/docs/public/execution/failures.mdx index 6884d5708..7113ac20a 100644 --- a/docs/public/execution/failures.mdx +++ b/docs/public/execution/failures.mdx @@ -121,7 +121,9 @@ provider = "anthropic" fallbacks = ["gemini", "openai"] ``` -When Anthropic is unavailable, Fabro tries Gemini first, then OpenAI. Each fallback entry may be a bare provider token (like `"gemini"`), a bare model alias (like `"gpt-5.4"`), or a qualified `"provider/model"` reference. For each fallback provider, Fabro selects the closest model by matching required capabilities (tool use, vision, reasoning) and minimizing cost difference. +When Anthropic is unavailable, Fabro tries Gemini first, then OpenAI. Each fallback entry may be a bare provider token (like `"gemini"`), a bare model alias (like `"gpt-5.4"`), or a qualified `"provider/model"` reference. Provider-only entries select the closest model by matching required capabilities (tool use, vision, reasoning) and minimizing cost difference. Qualified model entries are provider pins; bare models and aliases select among ready providers by priority. + +Fabro resolves the primary and fallback selectors to concrete provider/model offerings when it creates the run and persists the result. Resume reuses that materialized chain rather than re-ranking providers after credentials or priorities change. Runtime fallback is the mechanism for a provider failure that happens after creation. ### What triggers failover diff --git a/docs/public/execution/run-configuration.mdx b/docs/public/execution/run-configuration.mdx index 1b51a9161..20001cd86 100644 --- a/docs/public/execution/run-configuration.mdx +++ b/docs/public/execution/run-configuration.mdx @@ -138,11 +138,11 @@ name = "claude-sonnet-4-5" | Field | Description | |---|---| -| `name` | Model ID or alias (e.g. `claude-sonnet-4-5`, `opus`, `gemini-pro`). See [Models](/core-concepts/models). | -| `provider` | Provider name (optional — auto-inferred from the model catalog). Only needed for models not in the catalog or to force a specific provider. | -| `fallbacks` | Ordered list of model references to try when the primary is unavailable. Entries can be bare provider tokens (`"openai"`), bare model aliases, or qualified `"provider/model"` references. | +| `name` | Canonical model slug or alias (e.g. `claude-sonnet-4-5`, `opus`, `gemini-pro`). See [Models](/core-concepts/models). | +| `provider` | Optional provider pin. When omitted, Fabro selects a matching offering from ready providers by provider priority. When set, lookup is restricted to that provider and unavailability is an error. | +| `fallbacks` | Ordered list of model references to try after a failover-eligible error. Entries can be bare provider tokens (`"openai"`), bare model aliases, or qualified `"provider/model"` references. | -Provider values are catalog provider ID strings. Built-in IDs like `anthropic` and `openai` work, and settings-defined IDs like `proxy` work after they are added under `[llm.providers.]`. +Provider values are catalog provider ID strings. Built-in IDs like `anthropic` and `openai` work, and settings-defined IDs like `proxy` work after they are added under `[llm.providers.]`. A qualified fallback such as `"openrouter/gpt-56-sol"` is pinned to that provider; a bare alias can select among ready fallback offerings by priority. #### `[run.model.controls]` @@ -163,6 +163,12 @@ speed = "fast" | `reasoning_effort` | Native reasoning-effort value to request when the selected model allows it, such as `"low"`, `"medium"`, `"high"`, `"xhigh"`, or `"max"`. | | `speed` | Native speed value to request when the selected model declares it, such as `"fast"`. The standard speed is implicit and does not need to be set. | +#### Resolution and fallback behavior + +At run creation, Fabro resolves unqualified primary and fallback selectors to concrete provider and canonical-model pairs using the ready-provider snapshot, then persists those choices. Resume reuses the materialized routing and does not reconsider provider priority if credentials or configuration changed. Runtime failover walks the persisted fallback chain; create a new run to reselect from current provider availability. + +Provider-only fallbacks choose the closest compatible model on that provider. A provider-qualified model or alias resolves only within that provider, while a bare model or alias uses ready providers and priority. + #### Fallbacks with splice Use the reserved `"..."` marker in `fallbacks` to splice in the inherited list from lower-precedence layers: diff --git a/docs/public/integrations/litellm.mdx b/docs/public/integrations/litellm.mdx index 7f1891f8d..f7acf9fd4 100644 --- a/docs/public/integrations/litellm.mdx +++ b/docs/public/integrations/litellm.mdx @@ -24,24 +24,23 @@ _version = 1 enabled = true base_url = "http://localhost:4000/v1" -[llm.models."litellm-gpt-5"] -provider = "litellm" +[llm.providers.litellm.models."litellm-gpt-5"] api_id = "gpt-5" display_name = "LiteLLM GPT-5" family = "litellm" default = true -[llm.models."litellm-gpt-5".limits] +[llm.providers.litellm.models."litellm-gpt-5".limits] context_window = 128000 max_output = 8192 -[llm.models."litellm-gpt-5".features] +[llm.providers.litellm.models."litellm-gpt-5".features] tools = true vision = false reasoning = false ``` -`api_id` is the model name Fabro sends to LiteLLM. It should match a model name configured in your LiteLLM proxy. +`api_id` is the opaque model name Fabro sends to LiteLLM. It should match a model name configured in your LiteLLM proxy. When the provider-facing name is the same as the Fabro model slug, omit `api_id`; it defaults to the slug. ## Configure credentials diff --git a/lib/crates/fabro-config/src/builders.rs b/lib/crates/fabro-config/src/builders.rs index 1458ba0b9..78de24414 100644 --- a/lib/crates/fabro-config/src/builders.rs +++ b/lib/crates/fabro-config/src/builders.rs @@ -16,9 +16,9 @@ use crate::resolve::{ }; use crate::user::load_settings_config; use crate::{ - CliLayer, Combine, CostRates, EnvironmentLayer, Error, LlmLayer, LlmModelFeatures, - LlmModelLimits, MergeMap, ModelControls, ModelCostTable, ModelSettings, ProviderSettings, - Result, RunLayer, ServerLayer, SettingsLayer, run, + CliLayer, Combine, CostRates, EnvironmentLayer, Error, LegacyModelSettings, LlmLayer, + LlmModelFeatures, LlmModelLimits, MergeMap, ModelControls, ModelCostTable, ModelSettings, + ProviderSettings, Result, RunLayer, ServerLayer, SettingsLayer, run, }; #[derive(Debug, Clone, PartialEq, Eq)] @@ -321,7 +321,7 @@ fn llm_layer_to_catalog_settings(llm: LlmLayer) -> model_catalog::LlmCatalogSett .models .into_inner() .into_iter() - .map(|(id, settings)| (id, model_settings_to_catalog(settings))) + .map(|(id, settings)| (id, legacy_model_settings_to_catalog(settings))) .collect(), } } @@ -341,6 +341,12 @@ fn provider_settings_to_catalog( .collect() }); model_catalog::ProviderCatalogSettings { + models: settings + .models + .into_inner() + .into_iter() + .map(|(id, settings)| (id, model_settings_to_catalog(settings))) + .collect(), display_name: settings.display_name, adapter: settings.adapter, codec: settings.codec, @@ -356,9 +362,17 @@ fn provider_settings_to_catalog( } } +fn legacy_model_settings_to_catalog( + settings: LegacyModelSettings, +) -> model_catalog::ModelCatalogSettings { + let LegacyModelSettings { provider, model } = settings; + let mut settings = model_settings_to_catalog(model); + settings.provider = provider; + settings +} + fn model_settings_to_catalog(settings: ModelSettings) -> model_catalog::ModelCatalogSettings { let ModelSettings { - provider, api_id, codec, billing_policy, @@ -379,7 +393,7 @@ fn model_settings_to_catalog(settings: ModelSettings) -> model_catalog::ModelCat costs, } = settings; model_catalog::ModelCatalogSettings { - provider, + provider: None, api_id, codec, billing_policy, @@ -820,7 +834,7 @@ provider = "docker" } #[test] - fn server_runtime_settings_preserves_llm_catalog_overrides() { + fn server_runtime_settings_preserves_provider_scoped_llm_catalog_overrides() { let settings = server_runtime_settings_from_toml( r#" _version = 1 @@ -837,17 +851,16 @@ agent_profile = "anthropic" [llm.providers.acme.auth] credentials = ["env:ACME_API_KEY"] -[llm.models."acme-large"] -provider = "acme" +[llm.providers.acme.models."acme-large"] display_name = "Acme Large" family = "acme" default = true agent_profile = "gemini" -[llm.models."acme-large".limits] +[llm.providers.acme.models."acme-large".limits] context_window = 128000 -[llm.models."acme-large".features] +[llm.providers.acme.models."acme-large".features] tools = true vision = false reasoning = false @@ -857,20 +870,69 @@ reasoning = false ) .expect("server runtime settings should resolve"); - let catalog = - fabro_model::Catalog::from_builtin_with_overrides(&settings.llm_catalog_settings) - .expect("catalog overrides should build"); + let provider = settings + .llm_catalog_settings + .providers + .get("acme") + .expect("provider settings should be present"); + let model = provider + .models + .get("acme-large") + .expect("provider-scoped model settings should be present"); + assert_eq!(model.display_name.as_deref(), Some("Acme Large")); + assert_eq!(model.agent_profile, Some(fabro_model::AgentProfileKind::Gemini)); + assert!(settings.llm_catalog_settings.models.is_empty()); + } + + #[test] + fn server_runtime_settings_converts_legacy_models_to_provider_catalog_shape() { + let settings = server_runtime_settings_from_toml( + r#" +_version = 1 + +[server.auth] +methods = ["dev-token"] + +[llm.models."acme-large"] +provider = "acme" +display_name = "Acme Large" +"#, + None, + None, + ) + .expect("legacy catalog settings should resolve"); assert_eq!( - catalog - .get("acme-large") - .map(|model| model.provider.clone()), - Some(fabro_model::ProviderId::new("acme")) + settings.llm_catalog_settings.providers["acme"].models["acme-large"] + .display_name + .as_deref(), + Some("Acme Large") ); + assert!(settings.llm_catalog_settings.models.is_empty()); + } + + #[test] + fn server_runtime_settings_retains_providerless_legacy_models_for_catalog_adoption() { + let settings = server_runtime_settings_from_toml( + r#" +_version = 1 + +[server.auth] +methods = ["dev-token"] + +[llm.models."known-model"] +display_name = "Renamed Known Model" +"#, + None, + None, + ) + .expect("provider-less legacy catalog settings should resolve"); + assert_eq!( - catalog - .effective_agent_profile(&fabro_model::ProviderId::new("acme"), Some("acme-large")), - Some(fabro_model::AgentProfileKind::Gemini) + settings.llm_catalog_settings.models["known-model"] + .display_name + .as_deref(), + Some("Renamed Known Model") ); } diff --git a/lib/crates/fabro-config/src/layers/llm.rs b/lib/crates/fabro-config/src/layers/llm.rs index 5fb8a0a64..00fc79e35 100644 --- a/lib/crates/fabro-config/src/layers/llm.rs +++ b/lib/crates/fabro-config/src/layers/llm.rs @@ -12,11 +12,15 @@ //! enabled = true //! aliases = ["moonshot"] //! -//! [llm.models."kimi-k2.5"] -//! provider = "kimi" +//! [llm.providers.kimi.models."kimi-k2.5"] //! ... //! ``` //! +//! Legacy top-level `[llm.models.]` rows remain accepted by the settings +//! parser. Rows with a `provider` are normalized into the canonical provider +//! scope before layers combine; provider-less rows are retained for +//! catalog-aware compatibility handling. +//! //! Per-provider and per-model entries field-merge across layers (default → //! user → server → project → workflow/run). Inner arrays such as //! `auth.credentials`, `aliases`, `controls.reasoning_effort`, and @@ -43,15 +47,22 @@ pub struct LlmLayer { /// Provider definitions keyed by provider ID. #[serde(default, skip_serializing_if = "MergeMap::is_empty")] pub providers: MergeMap, - /// Model definitions keyed by canonical model ID. + /// Legacy top-level model definitions keyed by canonical model ID. + /// + /// Provider-qualified rows are moved into [`ProviderSettings::models`] by + /// the settings parser. Rows without a provider remain here until the + /// built-in catalog can adopt them unambiguously. #[serde(default, skip_serializing_if = "MergeMap::is_empty")] - pub models: MergeMap, + pub models: MergeMap, } /// One entry in `[llm.providers.]`. #[derive(Debug, Clone, Default, PartialEq, Serialize, Deserialize, fabro_macros::Combine)] #[serde(deny_unknown_fields)] pub struct ProviderSettings { + /// Model definitions owned by this provider, keyed by canonical model ID. + #[serde(default, skip_serializing_if = "MergeMap::is_empty")] + pub models: MergeMap, #[serde(default, skip_serializing_if = "Option::is_none")] pub display_name: Option, /// Adapter registry key (e.g. `"openai_compatible"`). @@ -89,13 +100,10 @@ pub struct ProviderSettings { pub aliases: Option>, } -/// One entry in `[llm.models.]`. +/// One entry in `[llm.providers..models.]`. #[derive(Debug, Clone, Default, PartialEq, Serialize, Deserialize, fabro_macros::Combine)] #[serde(deny_unknown_fields)] pub struct ModelSettings { - /// Provider ID this model belongs to. - #[serde(default, skip_serializing_if = "Option::is_none")] - pub provider: Option, /// Identifier sent to the provider API. Defaults to the catalog model ID /// when omitted. #[serde(default, skip_serializing_if = "Option::is_none")] @@ -155,6 +163,38 @@ pub struct ModelSettings { pub costs: Option, } +/// Input-only compatibility row for the legacy `[llm.models.]` shape. +#[derive(Debug, Clone, Default, PartialEq, Serialize, Deserialize, fabro_macros::Combine)] +#[serde(deny_unknown_fields)] +pub struct LegacyModelSettings { + /// Provider ID used to move this row into the canonical provider scope. + #[serde(default, skip_serializing_if = "Option::is_none")] + pub provider: Option, + #[serde(flatten)] + pub model: ModelSettings, +} + +impl LegacyModelSettings { + #[must_use] + pub(crate) fn into_model(self) -> ModelSettings { + self.model + } +} + +impl std::ops::Deref for LegacyModelSettings { + type Target = ModelSettings; + + fn deref(&self) -> &Self::Target { + &self.model + } +} + +impl std::ops::DerefMut for LegacyModelSettings { + fn deref_mut(&mut self) -> &mut Self::Target { + &mut self.model + } +} + #[derive(Debug, Clone, Default, PartialEq, Serialize, Deserialize, fabro_macros::Combine)] #[serde(deny_unknown_fields)] pub struct ModelLimits { diff --git a/lib/crates/fabro-config/src/layers/mod.rs b/lib/crates/fabro-config/src/layers/mod.rs index c3fa1632c..aae4c99a6 100644 --- a/lib/crates/fabro-config/src/layers/mod.rs +++ b/lib/crates/fabro-config/src/layers/mod.rs @@ -21,8 +21,8 @@ pub use environment::{ EnvironmentNetworkLayer, EnvironmentResourcesLayer, RunEnvironmentLayer, }; pub use llm::{ - CostRates, CredentialRef, CredentialRefParseError, LlmLayer, ModelControls, ModelCostTable, - ModelFeatures as LlmModelFeatures, ModelLimits as LlmModelLimits, ModelSettings, + CostRates, CredentialRef, CredentialRefParseError, LegacyModelSettings, LlmLayer, ModelControls, + ModelCostTable, ModelFeatures as LlmModelFeatures, ModelLimits as LlmModelLimits, ModelSettings, ProviderSettings, ReasoningEffortFeature, }; pub use log_filter::LogFilter; diff --git a/lib/crates/fabro-config/src/lib.rs b/lib/crates/fabro-config/src/lib.rs index f6097b9be..fd7d48608 100644 --- a/lib/crates/fabro-config/src/lib.rs +++ b/lib/crates/fabro-config/src/lib.rs @@ -46,8 +46,9 @@ pub use layers::{ CredentialRefParseError, EnvironmentDockerfileLayer, EnvironmentImageLayer, EnvironmentLayer, EnvironmentLifecycleLayer, EnvironmentNetworkLayer, EnvironmentResourcesLayer, GitAuthorLayer, GithubIntegrationLayer, HookAgentMarker, HookEntry, HookTlsMode, IntegrationWebhooksLayer, - InterviewProviderLayer, InterviewsLayer, LlmLayer, LlmModelFeatures, LlmModelLimits, LogFilter, - McpEntryLayer, MergeMap, ModelControls, ModelCostTable, ModelRefOrSplice, ModelSettings, + InterviewProviderLayer, InterviewsLayer, LegacyModelSettings, LlmLayer, LlmModelFeatures, + LlmModelLimits, LogFilter, McpEntryLayer, MergeMap, ModelControls, ModelCostTable, + ModelRefOrSplice, ModelSettings, NotificationProviderLayer, NotificationRouteLayer, ObjectStoreLocalLayer, ObjectStoreS3Layer, PrepareStep, ProjectLayer, ProviderSettings, ReasoningEffortFeature, ReplaceMap, RunAgentLayer, RunArtifactsLayer, RunCheckpointLayer, RunCloneLayer, RunEnvironmentLayer, RunExecutionLayer, diff --git a/lib/crates/fabro-config/src/parse.rs b/lib/crates/fabro-config/src/parse.rs index b4e008791..35b7f4709 100644 --- a/lib/crates/fabro-config/src/parse.rs +++ b/lib/crates/fabro-config/src/parse.rs @@ -39,6 +39,13 @@ pub enum ParseError { path: String, source: SettingsSource, }, + ConflictingLlmModelDefinitions { + provider: String, + model: String, + }, + InvalidLegacyLlmModelProvider { + model: String, + }, } impl fmt::Display for ParseError { @@ -60,6 +67,14 @@ impl fmt::Display for ParseError { f, "`{path}` is server-managed and cannot be set in {source} settings; configure cwd on a server-managed environment instead." ), + Self::ConflictingLlmModelDefinitions { provider, model } => write!( + f, + "model `{model}` on provider `{provider}` is defined in both `llm.models.{model}` and `llm.providers.{provider}.models.{model}` in the same settings source" + ), + Self::InvalidLegacyLlmModelProvider { model } => write!( + f, + "legacy model `llm.models.{model}` has an empty provider; omit it for catalog-aware adoption or set a non-empty provider ID" + ), } } } @@ -118,8 +133,36 @@ pub(crate) fn parse_settings(input: &str) -> Result { } } - raw.try_into::() - .map_err(|e| ParseError::Toml(e.to_string())) + let mut layer = raw + .try_into::() + .map_err(|e| ParseError::Toml(e.to_string()))?; + normalize_legacy_llm_models(&mut layer)?; + Ok(layer) +} + +fn normalize_legacy_llm_models(layer: &mut SettingsLayer) -> Result<(), ParseError> { + let Some(llm) = layer.llm.as_mut() else { + return Ok(()); + }; + + let legacy_models = std::mem::take(&mut llm.models).into_inner(); + for (model, legacy) in legacy_models { + let Some(provider) = legacy.provider.as_deref() else { + llm.models.insert(model, legacy); + continue; + }; + if provider.is_empty() { + return Err(ParseError::InvalidLegacyLlmModelProvider { model }); + } + + let provider = provider.to_string(); + let provider_settings = llm.providers.entry(provider.clone()).or_default(); + if provider_settings.models.contains_key(&model) { + return Err(ParseError::ConflictingLlmModelDefinitions { provider, model }); + } + provider_settings.models.insert(model, legacy.into_model()); + } + Ok(()) } #[derive(Debug, Clone, Copy, PartialEq, Eq)] @@ -289,11 +332,91 @@ mod tests { } #[test] - fn accepts_new_llm_models_subtree() { - let parsed = "[llm.models.\"foo\"]\nprovider = \"kimi\"\n" + fn accepts_provider_scoped_models_subtree() { + let parsed = "[llm.providers.kimi.models.\"foo\"]\ndisplay_name = \"Foo\"\n" .parse::() - .unwrap(); - assert!(parsed.llm.unwrap().models.contains_key("foo")); + .expect("provider-scoped model should parse"); + let llm = parsed.llm.expect("llm layer should be present"); + + assert_eq!( + llm.providers["kimi"].models["foo"].display_name.as_deref(), + Some("Foo") + ); + assert!(llm.models.is_empty()); + } + + #[test] + fn normalizes_legacy_llm_model_with_provider_into_provider_scope() { + let parsed = r#" +[llm.models.foo] +provider = "kimi" +display_name = "Foo" +"# + .parse::() + .expect("legacy model should parse"); + let llm = parsed.llm.expect("llm layer should be present"); + + assert_eq!( + llm.providers["kimi"].models["foo"].display_name.as_deref(), + Some("Foo") + ); + assert!(llm.models.is_empty()); + } + + #[test] + fn retains_providerless_legacy_llm_model_for_catalog_aware_adoption() { + let parsed = r#" +[llm.models.foo] +display_name = "Renamed Foo" +"# + .parse::() + .expect("provider-less legacy model should remain compatible"); + let llm = parsed.llm.expect("llm layer should be present"); + + assert_eq!( + llm.models["foo"].display_name.as_deref(), + Some("Renamed Foo") + ); + assert!(llm.providers.is_empty()); + } + + #[test] + fn rejects_same_source_legacy_and_provider_scoped_model_pair() { + let error = r#" +[llm.providers.kimi.models.foo] +display_name = "Canonical Foo" + +[llm.models.foo] +provider = "kimi" +display_name = "Legacy Foo" +"# + .parse::() + .expect_err("same pair in both syntaxes should be rejected"); + + assert_eq!( + error, + ParseError::ConflictingLlmModelDefinitions { + provider: "kimi".to_string(), + model: "foo".to_string(), + } + ); + } + + #[test] + fn rejects_empty_legacy_llm_model_provider_with_typed_error() { + let error = r#" +[llm.models.foo] +provider = "" +"# + .parse::() + .expect_err("empty legacy provider should be rejected"); + + assert_eq!( + error, + ParseError::InvalidLegacyLlmModelProvider { + model: "foo".to_string(), + } + ); } #[test] diff --git a/lib/crates/fabro-config/src/tests/combine.rs b/lib/crates/fabro-config/src/tests/combine.rs index ff95a8aec..2a6c487e7 100644 --- a/lib/crates/fabro-config/src/tests/combine.rs +++ b/lib/crates/fabro-config/src/tests/combine.rs @@ -318,3 +318,117 @@ bucket = "higher-bucket" assert_eq!(s3.bucket, Some("higher-bucket".to_string())); assert_eq!(s3.region, None); } + +#[test] +fn provider_and_model_rows_field_merge_independently() { + let lower = parse( + r#" +[llm.providers.acme] +display_name = "Acme" +adapter = "openai_compatible" +base_url = "https://lower.example/v1" + +[llm.providers.acme.models.large] +display_name = "Acme Large" +family = "acme" + +[llm.providers.acme.models.large.limits] +context_window = 128000 +max_output = 32000 +"#, + ); + let higher = parse( + r#" +[llm.providers.acme] +base_url = "https://higher.example/v1" + +[llm.providers.acme.models.large] +display_name = "Acme Large v2" + +[llm.providers.acme.models.large.limits] +max_output = 64000 +"#, + ); + + let merged = higher.combine(lower); + let acme = &merged.llm.expect("llm layer should be present").providers["acme"]; + assert_eq!(acme.display_name.as_deref(), Some("Acme")); + assert_eq!(acme.adapter.as_deref(), Some("openai_compatible")); + assert_eq!(acme.base_url.as_deref(), Some("https://higher.example/v1")); + + let model = &acme.models["large"]; + assert_eq!(model.display_name.as_deref(), Some("Acme Large v2")); + assert_eq!(model.family.as_deref(), Some("acme")); + assert_eq!( + model.limits.as_ref().and_then(|limits| limits.context_window), + Some(128_000) + ); + assert_eq!( + model.limits.as_ref().and_then(|limits| limits.max_output), + Some(64_000) + ); +} + +#[test] +fn legacy_model_is_normalized_before_cross_source_combine() { + let lower = parse( + r#" +[llm.models.large] +provider = "acme" +family = "acme" + +[llm.models.large.limits] +context_window = 128000 +"#, + ); + let higher = parse( + r#" +[llm.providers.acme.models.large] +display_name = "Acme Large" + +[llm.providers.acme.models.large.limits] +max_output = 64000 +"#, + ); + + let merged = higher.combine(lower); + let llm = merged.llm.expect("llm layer should be present"); + let model = &llm.providers["acme"].models["large"]; + + assert_eq!(model.display_name.as_deref(), Some("Acme Large")); + assert_eq!(model.family.as_deref(), Some("acme")); + assert_eq!( + model.limits.as_ref().and_then(|limits| limits.context_window), + Some(128_000) + ); + assert_eq!( + model.limits.as_ref().and_then(|limits| limits.max_output), + Some(64_000) + ); + assert!(llm.models.is_empty()); +} + +#[test] +fn same_model_id_on_different_providers_stays_independent() { + let merged = parse( + r#" +[llm.providers.openai.models.shared] +api_id = "shared" + +[llm.providers.openrouter.models.shared] +api_id = "openai/shared" +"#, + ); + let llm = merged.llm.expect("llm layer should be present"); + + assert_eq!( + llm.providers["openai"].models["shared"].api_id.as_deref(), + Some("shared") + ); + assert_eq!( + llm.providers["openrouter"].models["shared"] + .api_id + .as_deref(), + Some("openai/shared") + ); +} diff --git a/lib/crates/fabro-dev/src/commands/docs_options_reference.rs b/lib/crates/fabro-dev/src/commands/docs_options_reference.rs index aacc08575..a446d048a 100644 --- a/lib/crates/fabro-dev/src/commands/docs_options_reference.rs +++ b/lib/crates/fabro-dev/src/commands/docs_options_reference.rs @@ -247,19 +247,20 @@ x-team-secret = "{{ secrets.gateway_team_secret }}" | `auth.credentials` | array | required when `auth` present | Ordered credential refs. Accepted forms are `vault:`, `env:`, and `aws_sigv4` (sign requests from the AWS default credential chain — Bedrock). Literal secret strings are rejected. | | `auth.header` | `"bearer"` or `{ custom = "Header-Name" }` | `"bearer"` | Primary API-key header policy. Omit when the provider uses a standard bearer token. | | `extra_headers` | table | `{}` | Additional headers attached to provider requests. Values are interpolation strings: literal text, an `{{ env.NAME }}` token, or a `{{ secrets.NAME }}` token. Put credentials in a secret and reference them with a `{{ secrets.NAME }}` token, not a bare literal. | -| `priority` | integer | `0` | Higher-priority configured providers win default selection; ties use canonical provider ID. | +| `priority` | integer | `0` | Higher-priority ready providers win unqualified model selection; ties use canonical provider ID in ascending order. | | `enabled` | boolean | `true` | Set `false` to disable a provider after lower-precedence layers define it. | | `aliases` | array | `[]` | Additional provider names accepted by model routing and fallback config. | -## `[llm.models.]` +## `[llm.providers..models.]` -Define or override a model in the catalog. The table key is the canonical -model ID Fabro users reference; `api_id` is the model string sent to the -provider API. +Define or override one provider-specific model offering. The containing table +supplies the provider ID, and `` is the canonical, human-facing model +slug used by workflows. The stable identity of an offering is the pair +`(provider, model)`; the same model slug and aliases may be reused by other +providers. ```toml title="settings.toml" -[llm.models."team-code-large"] -provider = "proxy" +[llm.providers.proxy.models."team-code-large"] api_id = "provider-wire-model-name" agent_profile = "anthropic" display_name = "Team Code Large" @@ -270,56 +271,66 @@ enabled = true aliases = ["team-code"] estimated_output_tps = 80 -[llm.models."team-code-large".limits] +[llm.providers.proxy.models."team-code-large".limits] context_window = 200000 max_output = 32000 -[llm.models."team-code-large".features] +[llm.providers.proxy.models."team-code-large".features] tools = true vision = false reasoning = true reasoning_effort = "levels" prompt_cache = true -[llm.models."team-code-large".controls] +[llm.providers.proxy.models."team-code-large".controls] reasoning_effort = ["low", "medium", "high"] speed = ["fast"] -[llm.models."team-code-large".costs] +[llm.providers.proxy.models."team-code-large".costs] input_cost_per_mtok = 1.50 output_cost_per_mtok = 8.00 cache_input_cost_per_mtok = 0.30 -[llm.models."team-code-large".costs.speed.fast] +[llm.providers.proxy.models."team-code-large".costs.speed.fast] input_cost_per_mtok = 3.00 output_cost_per_mtok = 16.00 cache_input_cost_per_mtok = 0.60 ``` +`api_id` is an opaque provider wire identifier, not a workflow selector or a +routing namespace. When omitted, it defaults to the exact canonical model +slug. Set it only when the provider expects a different value; an explicitly +empty value is invalid. + +Unqualified model selectors consider ready providers and then choose the +highest provider `priority`, with canonical provider ID as the deterministic +tie-breaker. Supplying a provider pins lookup to that provider. An alias must +identify only one model within a provider, but reusing it on another provider +is valid and enables portable workflow selectors. + | Key | Type / values | Default | Description | |---|---|---|---| -| `provider` | string | None | Provider ID this model belongs to. | -| `api_id` | string | model ID | Identifier sent to the provider API. | +| `api_id` | string | canonical model slug | Opaque identifier sent to this offering's provider API. It is not parsed for routing. | | `agent_profile` | `"anthropic"` \| `"openai"` \| `"gemini"` | provider profile | Agent profile override for this model. Model overrides take precedence over provider overrides. | | `billing_policy` | `"openai"` \| `"anthropic"` \| `"gemini"` \| `"none"` | provider policy | Billing algorithm override for this model — for models whose billing family differs from their provider's (e.g. Claude served through OpenRouter bills Anthropic-style cache reads/writes). | -| `display_name` | string | model ID | Human-readable model name. | -| `family` | string | model ID | Family label used for catalog display and matching. | +| `display_name` | string | model slug | Human-readable model name. | +| `family` | string | model slug | Family metadata used for catalog display and matching; it is not a routing namespace. | | `training` | string | None | Training data cutoff label. | | `knowledge_cutoff` | string or TOML date | None | Public knowledge cutoff label; TOML dates normalize to `YYYY-MM-DD`. | | `default` | boolean | `false` | Whether this is the provider default model. | | `probe` | boolean | `false` | Whether this model should be preferred for provider connectivity probes. Set `false` in a higher-precedence layer to clear an inherited probe marker. | | `enabled` | boolean | `true` | Set `false` to disable a model after lower-precedence layers define it. | -| `aliases` | array | `[]` | Additional model names accepted by routing and fallback config. | +| `aliases` | array | `[]` | Additional user-facing selectors. Each selector must be unique within this provider but may be reused by other providers. | | `estimated_output_tps` | number | None | Estimated output tokens per second for catalog display and planning. | -## `[llm.models..limits]` +## `[llm.providers..models..limits]` | Key | Type / values | Default | Description | |---|---|---|---| | `context_window` | integer | None | Maximum context window size in tokens. | | `max_output` | integer | None | Maximum output tokens, if known. | -## `[llm.models..features]` +## `[llm.providers..models..features]` | Key | Type / values | Default | Description | |---|---|---|---| @@ -330,14 +341,14 @@ cache_input_cost_per_mtok = 0.60 | `prompt_cache` | boolean | `false` | Whether prompt cache pricing/usage applies. | | `sampling_params` | boolean | `true` | Whether the model accepts classic sampling parameters (`temperature`, `top_p`). | -## `[llm.models..controls]` +## `[llm.providers..models..controls]` | Key | Type / values | Default | Description | |---|---|---|---| | `reasoning_effort` | array | all standard levels when feature is `"levels"` or `"always_adaptive"` | User-facing reasoning effort values Fabro may send for this model. Can be set explicitly for reasoning models whose provider adapter maps effort to a non-native API shape. | | `speed` | array | `[]` | Additional speeds beyond implicit `standard`; do not list `standard`. | -## `[llm.models..costs]` +## `[llm.providers..models..costs]` | Key | Type / values | Default | Description | |---|---|---|---| @@ -345,11 +356,12 @@ cache_input_cost_per_mtok = 0.60 | `output_cost_per_mtok` | number | None | Output cost in USD per million tokens. | | `cache_input_cost_per_mtok` | number | None | Cached input/read cost in USD per million tokens. | -## `[llm.models..costs.speed.]` +## `[llm.providers..models..costs.speed.]` -Per-speed cost overrides use the same keys as `[llm.models..costs]`. -Each `` key must be declared in `[llm.models..controls].speed`. -The `standard` speed is implicit and always uses the base cost table. +Per-speed cost overrides use the same keys as +`[llm.providers..models..costs]`. Each `` key must be +declared in `[llm.providers..models..controls].speed`. The +`standard` speed is implicit and always uses the base cost table. "#, ); @@ -389,3 +401,21 @@ See [MCP](/agents/mcp) for transport-specific examples. fn normalize_doc(doc: &str) -> String { doc.trim().trim_end_matches('.').to_string() } + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn llm_catalog_reference_teaches_provider_scoped_portable_models() { + let reference = render_options_reference(); + + assert!(reference.contains("## `[llm.providers..models.]`")); + assert!(reference.contains("[llm.providers.proxy.models.\"team-code-large\"]")); + assert!(reference.contains("defaults to the exact canonical model\nslug")); + assert!(reference.contains("Supplying a provider pins lookup to that provider")); + assert!(reference.contains("reusing it on another provider\nis valid")); + assert!(reference.contains("opaque provider wire identifier")); + assert!(!reference.contains("## `[llm.models.]`")); + } +} diff --git a/lib/crates/fabro-llm/src/client.rs b/lib/crates/fabro-llm/src/client.rs index c72e9ea50..7fe5011e2 100644 --- a/lib/crates/fabro-llm/src/client.rs +++ b/lib/crates/fabro-llm/src/client.rs @@ -597,6 +597,7 @@ fn format_additional_speeds(values: &[Speed]) -> String { #[cfg(test)] mod tests { + use std::sync::Mutex; use std::sync::atomic::{AtomicUsize, Ordering}; use async_trait::async_trait; diff --git a/lib/crates/fabro-llm/src/cost.rs b/lib/crates/fabro-llm/src/cost.rs index e03ead5de..00371919c 100644 --- a/lib/crates/fabro-llm/src/cost.rs +++ b/lib/crates/fabro-llm/src/cost.rs @@ -25,11 +25,11 @@ pub(crate) fn estimate_cost_usd( let catalog = catalog?; // The billing machinery compares ModelRefs against the catalog's // canonical identity, so resolve model aliases and provider names first. - let model = catalog.get(model)?; let provider = catalog.provider(&ProviderId::new(provider))?; + let model = catalog.model_on_provider(&provider.id, model)?; let model_ref = ModelRef { provider: provider.id.clone(), - model_id: model.id.clone(), + model_id: model.id.to_string(), speed, }; let micros = catalog.price_tokens(&model_ref, tokens)?; diff --git a/lib/crates/fabro-llm/src/model_test.rs b/lib/crates/fabro-llm/src/model_test.rs index 47bca353c..09dc1e632 100644 --- a/lib/crates/fabro-llm/src/model_test.rs +++ b/lib/crates/fabro-llm/src/model_test.rs @@ -132,7 +132,7 @@ fn build_deep_test_params(info: &Model, client: Arc) -> Option) -> ModelRef { ModelRef { provider: self.provider.clone(), - model_id: self.id.clone(), + model_id: self.id.to_string(), speed, } } @@ -544,7 +544,7 @@ fn pricing_for_model_costs( Some(ModelPricing { model: ModelRef { provider: provider_id, - model_id: model.id.clone(), + model_id: model.id.to_string(), speed, }, policy, diff --git a/lib/crates/fabro-model/src/catalog.rs b/lib/crates/fabro-model/src/catalog.rs index 9ae07fc8b..0cc3fa0a8 100644 --- a/lib/crates/fabro-model/src/catalog.rs +++ b/lib/crates/fabro-model/src/catalog.rs @@ -12,7 +12,7 @@ use tracing::warn; use crate::Speed; use crate::adapter::{AdapterKind, AgentProfileKind}; use crate::codec::CodecKind; -use crate::ids::ProviderId; +use crate::ids::{ModelId, ProviderId}; use crate::provider::Provider; use crate::reasoning::ReasoningEffort; use crate::types::{Model, ModelCosts, ModelFeatures, ModelLimits, ReasoningEffortFeature}; @@ -39,6 +39,9 @@ pub struct LlmCatalogSettings { #[derive(Debug, Clone, Default, PartialEq, Deserialize)] #[serde(deny_unknown_fields)] pub struct ProviderCatalogSettings { + /// Provider-scoped model rows keyed by canonical human-facing model slug. + #[serde(default)] + pub models: HashMap, #[serde(default)] pub display_name: Option, #[serde(default)] @@ -382,6 +385,16 @@ static GLOBAL_CATALOG: LazyLock = LazyLock::new(|| { Catalog::from_builtin_toml().expect("embedded provider TOML files must build a valid catalog") }); +/// A built-in model identifier that was replaced by a provider-scoped model +/// slug. Retired identifiers are errors rather than aliases: silently accepting +/// one could route a persisted reference to a different provider. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct RetiredModelIdentifier { + pub identifier: String, + pub provider: ProviderId, + pub model: ModelId, +} + /// A resolved fallback target: provider name + model ID. #[derive(Debug, Clone, PartialEq, Eq)] pub struct FallbackTarget { @@ -528,12 +541,24 @@ pub enum CatalogBuildError { model: String, provider: ProviderId, }, - #[error("model identifier '{identifier}' is declared by both '{first}' and '{second}'")] - DuplicateModelIdentifier { + #[error( + "provider '{provider}' model identifier '{identifier}' is declared by both '{first}' and '{second}'" + )] + DuplicateProviderModelIdentifier { + provider: ProviderId, identifier: String, - first: String, - second: String, + first: ModelId, + second: ModelId, }, + #[error("provider '{provider}' model '{model}' configures an empty api_id")] + EmptyModelApiId { + provider: ProviderId, + model: ModelId, + }, + #[error( + "legacy model row '{model}' does not name a provider and does not uniquely match a built-in offering" + )] + AmbiguousLegacyModelProvider { model: String }, #[error("provider '{provider}' has multiple default models: {models:?}")] MultipleProviderDefaults { provider: ProviderId, @@ -574,17 +599,44 @@ pub enum CatalogBuildError { UndeclaredSpeedCost { model: String, speed: Speed }, } +/// Failure to select one offering for a user-facing model selector. +#[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)] +pub enum ModelSelectionError { + #[error( + "model identifier '{identifier}' was retired; use provider '{provider}' with model '{model}'" + )] + RetiredIdentifier { + identifier: String, + provider: ProviderId, + model: ModelId, + }, + #[error("unknown model selector '{selector}'")] + UnknownSelector { selector: String }, + #[error("model selector '{selector}' has no offering on an eligible provider")] + NoEligibleOffering { selector: String }, + #[error("provider '{provider}' is not available")] + UnavailableProvider { provider: ProviderId }, + #[error("provider '{provider}' has no model matching selector '{selector}'")] + UnknownSelectorOnProvider { + provider: ProviderId, + selector: String, + }, +} + /// Typed model catalog backed by a `Vec`. /// /// Use [`Catalog::builtin()`] for the embedded settings-backed catalog. #[derive(Debug)] pub struct Catalog { - models: Vec, - providers: Vec, - model_settings: HashMap, - model_index: HashMap, - provider_aliases: HashMap, - provider_index: HashMap, + models: Vec, + providers: Vec, + model_settings: HashMap<(ProviderId, ModelId), CatalogModelSettings>, + offering_index: HashMap<(ProviderId, ModelId), usize>, + canonical_candidates: HashMap>, + alias_candidates: HashMap>, + provider_aliases: HashMap, + provider_index: HashMap, + retired_identifiers: HashMap, } impl Catalog { @@ -617,27 +669,25 @@ impl Catalog { .collect(); let mut models_with_settings = Vec::new(); - let mut model_identifiers = BTreeMap::::new(); + let mut model_identifiers = HashMap::>::new(); let mut defaults_by_provider = HashMap::>::new(); let mut small_defaults_by_provider = HashMap::>::new(); - let mut model_ids = settings.models.keys().cloned().collect::>(); - model_ids.sort_unstable(); - for model_id in model_ids { - let model_settings = settings - .models - .get(&model_id) - .expect("model ID came from settings map keys"); + let normalized_models = normalized_model_settings(settings)?; + let mut model_keys = normalized_models.keys().cloned().collect::>(); + model_keys.sort_unstable(); + for (provider_id, model_id) in model_keys { + let model_settings = normalized_models + .get(&(provider_id.clone(), model_id.clone())) + .expect("model key came from normalized settings"); if model_settings.enabled == Some(false) { continue; } - let provider_id = - required_model_string(&model_id, model_settings.provider.as_ref(), "provider")?; if !known_providers.contains(provider_id.as_str()) { return Err(CatalogBuildError::UnknownModelProvider { - model: model_id, - provider: ProviderId::from(provider_id), + model: model_id, + provider: provider_id, }); } if !enabled_providers.contains(provider_id.as_str()) { @@ -649,22 +699,33 @@ impl Catalog { .expect("enabled provider ID should have provider metadata"); let (model, resolved_settings) = build_model(&model_id, model_settings, provider)?; - register_model_identifier(&mut model_identifiers, model.id.clone(), model.id.clone())?; + let provider_identifiers = model_identifiers.entry(provider_id.clone()).or_default(); + register_model_identifier( + provider_identifiers, + model.id.as_str().to_string(), + model.id.clone(), + &provider_id, + )?; for alias in &model.aliases { - register_model_identifier(&mut model_identifiers, alias.clone(), model.id.clone())?; + register_model_identifier( + provider_identifiers, + alias.clone(), + model.id.clone(), + &provider_id, + )?; } if model.default { defaults_by_provider .entry(model.provider.clone()) .or_default() - .push(model.id.clone()); + .push(model.id.to_string()); } if model.small_default { small_defaults_by_provider .entry(model.provider.clone()) .or_default() - .push(model.id.clone()); + .push(model.id.to_string()); } models_with_settings.push((model, resolved_settings)); } @@ -691,21 +752,25 @@ impl Catalog { models_with_settings.sort_by(|(left, _), (right, _)| model_order(left, right)); warn_multiple_probe_models(&models_with_settings); - let mut model_settings_by_id = HashMap::new(); + let mut model_settings = HashMap::new(); let mut models = Vec::new(); for (model, settings) in models_with_settings { - model_settings_by_id.insert(model.id.clone(), settings); + model_settings.insert((model.provider.clone(), model.id.clone()), settings); models.push(model); } - let model_index = build_model_index(&models); + let (offering_index, canonical_candidates, alias_candidates) = + build_model_indexes(&models); Ok(Self { models, providers, - model_settings: model_settings_by_id, - model_index, + model_settings, + offering_index, + canonical_candidates, + alias_candidates, provider_aliases, provider_index, + retired_identifiers: HashMap::new(), }) } @@ -713,8 +778,11 @@ impl Catalog { overrides: &LlmCatalogSettings, ) -> Result { let builtins = Self::builtin_settings()?; - let settings = merge_catalog_settings(overrides.clone(), builtins); - Self::from_settings(&settings) + let overrides = adopt_legacy_models(overrides.clone(), &builtins)?; + let settings = merge_catalog_settings(overrides, builtins); + let mut catalog = Self::from_settings(&settings)?; + catalog.retired_identifiers = builtin_retired_identifiers(); + Ok(catalog) } /// Builds a fresh catalog from embedded provider TOML without user @@ -749,7 +817,13 @@ impl Catalog { source, })?; validate_builtin_fragment(&path, &fragment)?; - layer.providers.extend(fragment.providers); + for (id, provider) in fragment.providers { + let provider = match layer.providers.remove(&id) { + Some(existing) => merge_provider_settings(provider, existing), + None => provider, + }; + layer.providers.insert(id, provider); + } layer.models.extend(fragment.models); } @@ -757,17 +831,116 @@ impl Catalog { } fn from_builtin_toml() -> Result { - Self::from_settings(&Self::builtin_settings()?) + let mut catalog = Self::from_settings(&Self::builtin_settings()?)?; + catalog.retired_identifiers = builtin_retired_identifiers(); + Ok(catalog) } - /// Look up a model by ID or alias. + /// Look up a selector across all enabled providers using catalog provider + /// priority. Runtime callers should prefer [`Self::select_model`] and pass + /// their ready-provider set explicitly. #[must_use] - pub fn get(&self, id: &str) -> Option<&Model> { - self.model_index - .get(id) + pub fn get(&self, selector: &str) -> Option<&Model> { + self.select_candidates(selector) + .and_then(|candidates| candidates.first()) .and_then(|idx| self.models.get(*idx)) } + /// Resolve a canonical model ID or alias only on the named provider. + #[must_use] + pub fn model_on_provider( + &self, + provider_id: &ProviderId, + selector: &str, + ) -> Option<&Model> { + let provider = self.provider(provider_id)?; + if let Some(idx) = self + .offering_index + .get(&(provider.id.clone(), ModelId::new(selector))) + { + return self.models.get(*idx); + } + self.alias_candidates + .get(selector)? + .iter() + .filter_map(|idx| self.models.get(*idx)) + .find(|model| model.provider == provider.id) + } + + /// Return the replacement address for a retired built-in identifier. + #[must_use] + pub fn retired_identifier(&self, identifier: &str) -> Option { + let (provider, model) = self.retired_identifiers.get(identifier)?; + Some(RetiredModelIdentifier { + identifier: identifier.to_string(), + provider: provider.clone(), + model: model.clone(), + }) + } + + /// Select one offering for a selector from caller-supplied eligible + /// providers. Canonical ID candidates are considered before aliases; + /// candidates are pre-sorted by provider priority descending and canonical + /// provider ID ascending. + pub fn select_model( + &self, + selector: &str, + explicit_provider: Option<&ProviderId>, + eligible_providers: &[ProviderId], + ) -> Result<&Model, ModelSelectionError> { + if let Some(retired) = self.retired_identifier(selector) { + return Err(ModelSelectionError::RetiredIdentifier { + identifier: retired.identifier, + provider: retired.provider, + model: retired.model, + }); + } + + let eligible = eligible_providers + .iter() + .filter_map(|id| self.provider(id).map(|provider| provider.id.clone())) + .collect::>(); + + if let Some(explicit_provider) = explicit_provider { + let provider = self + .provider(explicit_provider) + .ok_or_else(|| ModelSelectionError::UnknownSelectorOnProvider { + provider: explicit_provider.clone(), + selector: selector.to_string(), + })?; + if !eligible.contains(&provider.id) { + return Err(ModelSelectionError::UnavailableProvider { + provider: provider.id.clone(), + }); + } + return self.model_on_provider(&provider.id, selector).ok_or_else(|| { + ModelSelectionError::UnknownSelectorOnProvider { + provider: provider.id.clone(), + selector: selector.to_string(), + } + }); + } + + let candidates = self + .select_candidates(selector) + .ok_or_else(|| ModelSelectionError::UnknownSelector { + selector: selector.to_string(), + })?; + candidates + .iter() + .filter_map(|idx| self.models.get(*idx)) + .find(|model| eligible.contains(&model.provider)) + .ok_or_else(|| ModelSelectionError::NoEligibleOffering { + selector: selector.to_string(), + }) + } + + fn select_candidates(&self, selector: &str) -> Option<&Vec> { + self.canonical_candidates + .get(selector) + .or_else(|| self.alias_candidates.get(selector)) + } + #[must_use] pub fn providers(&self) -> &[CatalogProvider] { &self.providers @@ -786,7 +959,7 @@ impl Catalog { let stats = stats_by_provider.entry(model.provider.clone()).or_default(); stats.model_count = stats.model_count.saturating_add(1); if model.default { - stats.default_model = Some(model.id.clone()); + stats.default_model = Some(model.id.to_string()); } } @@ -817,10 +990,29 @@ impl Catalog { self.provider(id)?.vault_secret_name() } + /// Resolve settings for the highest-priority offering matching a selector. + /// Prefer [`Self::model_settings_on_provider`] or + /// [`Self::model_settings_for`] when provider identity is known. #[must_use] - pub fn model_settings(&self, id: &str) -> Option<&CatalogModelSettings> { - let model = self.get(id)?; - self.model_settings.get(&model.id) + pub fn model_settings(&self, selector: &str) -> Option<&CatalogModelSettings> { + let model = self.get(selector)?; + self.model_settings_for(model) + } + + #[must_use] + pub fn model_settings_on_provider( + &self, + provider: &ProviderId, + selector: &str, + ) -> Option<&CatalogModelSettings> { + let model = self.model_on_provider(provider, selector)?; + self.model_settings_for(model) + } + + #[must_use] + pub fn model_settings_for(&self, model: &Model) -> Option<&CatalogModelSettings> { + self.model_settings + .get(&(model.provider.clone(), model.id.clone())) } #[must_use] @@ -831,9 +1023,8 @@ impl Catalog { ) -> Option { let provider = self.provider(provider_id)?; let model_profile = model_id_or_alias - .and_then(|model_id| self.get(model_id)) - .filter(|model| model.provider == provider.id) - .and_then(|model| self.model_settings.get(&model.id)) + .and_then(|model_id| self.model_on_provider(&provider.id, model_id)) + .and_then(|model| self.model_settings_for(model)) .map(|settings| settings.agent_profile); Some(model_profile.unwrap_or(provider.agent_profile)) } @@ -849,9 +1040,8 @@ impl Catalog { ) -> Option { let provider = self.provider(provider_id)?; let model_codec = model_id_or_alias - .and_then(|model_id| self.get(model_id)) - .filter(|model| model.provider == provider.id) - .and_then(|model| self.model_settings.get(&model.id)) + .and_then(|model_id| self.model_on_provider(&provider.id, model_id)) + .and_then(|model| self.model_settings_for(model)) .map(|settings| settings.codec); Some(model_codec.unwrap_or(provider.codec)) } @@ -867,9 +1057,8 @@ impl Catalog { ) -> Option { let provider = self.provider(provider_id)?; let model_policy = model_id_or_alias - .and_then(|model_id| self.get(model_id)) - .filter(|model| model.provider == provider.id) - .and_then(|model| self.model_settings.get(&model.id)) + .and_then(|model_id| self.model_on_provider(&provider.id, model_id)) + .and_then(|model| self.model_settings_for(model)) .map(|settings| settings.billing_policy); Some(model_policy.unwrap_or(provider.billing_policy)) } @@ -993,8 +1182,7 @@ impl Catalog { if let Some(model) = self.models.iter().find(|model| { &model.provider == provider_id && self - .model_settings - .get(&model.id) + .model_settings_for(model) .is_some_and(|settings| settings.probe) }) { return Some(model); @@ -1043,7 +1231,7 @@ impl Catalog { model: &str, fallbacks: &HashMap>, ) -> Vec { - let Some(reference) = self.get(model) else { + let Some(reference) = self.model_on_provider(primary, model) else { return Vec::new(); }; @@ -1057,22 +1245,250 @@ impl Catalog { let provider = ProviderId::from(provider_str.clone()); self.closest(&provider, reference).map(|m| FallbackTarget { provider: provider_str.clone(), - model: m.id.clone(), + model: m.id.to_string(), }) }) .collect() } } -fn build_model_index(models: &[Model]) -> HashMap { - let mut index = HashMap::new(); +type OfferingIndex = HashMap<(ProviderId, ModelId), usize>; +type CanonicalCandidates = HashMap>; +type AliasCandidates = HashMap>; + +fn builtin_retired_identifiers() -> HashMap { + const RETIRED: &[(&str, &str, &str)] = &[ + ("openai.gpt-5.5", "bedrock-openai", "gpt-5.5"), + ("openai.gpt-5.4", "bedrock-openai", "gpt-5.4"), + ( + "us.anthropic.claude-sonnet-4-6", + "bedrock", + "claude-sonnet-4-6", + ), + ( + "us.anthropic.claude-opus-4-8", + "bedrock", + "claude-opus-4-8", + ), + ( + "us.anthropic.claude-haiku-4-5", + "bedrock", + "claude-haiku-4-5", + ), + ("openai.gpt-oss-120b", "bedrock", "gpt-oss-120b"), + ("openai.gpt-oss-20b", "bedrock", "gpt-oss-20b"), + ("amazon.nova-2-lite", "bedrock", "nova-2-lite"), + ("meta.llama4-maverick", "bedrock", "llama-4-maverick"), + ( + "mistral.mistral-large-3", + "bedrock", + "mistral-large-3", + ), + ("mistral.devstral-2", "bedrock", "devstral-2"), + ("deepseek.v3-2", "bedrock", "deepseek-v3.2"), + ("moonshotai.kimi-k2.5", "bedrock", "kimi-k2.5"), + ("zai.glm-5", "bedrock", "glm-5"), + ("minimax.minimax-m2.5", "bedrock", "minimax-m2.5"), + ( + "nvidia.nemotron-3-super", + "bedrock", + "nemotron-3-super", + ), + ( + "us.anthropic.claude-fable-5", + "bedrock", + "claude-fable-5", + ), + ( + "anthropic/claude-opus-4-7", + "openrouter", + "claude-opus-4-7", + ), + ( + "anthropic/claude-sonnet-4-6", + "openrouter", + "claude-sonnet-4-6", + ), + ( + "anthropic/claude-haiku-4-5", + "openrouter", + "claude-haiku-4-5", + ), + ("openai/gpt-5.4", "openrouter", "gpt-5.4"), + ("openai/gpt-5.5", "openrouter", "gpt-5.5"), + ( + "google/gemini-3.1-pro-preview", + "openrouter", + "gemini-3.1-pro-preview", + ), + ( + "google/gemini-3.5-flash", + "openrouter", + "gemini-3.5-flash", + ), + ("xiaomi/mimo-v2.5-pro", "openrouter", "mimo-v2.5-pro"), + ( + "minimax/minimax-m2.7", + "openrouter", + "minimax-m2.7", + ), + ( + "deepseek/deepseek-v4-pro", + "openrouter", + "deepseek-v4-pro", + ), + ( + "deepseek/deepseek-v4-flash", + "openrouter", + "deepseek-v4-flash", + ), + ("moonshotai/kimi-k2.6", "openrouter", "kimi-k2.6"), + ("moonshotai/kimi-k3", "openrouter", "kimi-k3"), + ( + "poolside/laguna-s-2.1", + "openrouter", + "laguna-s-2.1", + ), + ( + "poolside/laguna-xs-2.1", + "openrouter", + "laguna-xs-2.1", + ), + ("qwen/qwen3-coder", "openrouter", "qwen3-coder"), + ("qwen/qwen3.6-flash", "openrouter", "qwen3.6-flash"), + ("z-ai/glm-5.2", "openrouter", "glm-5.2"), + ("z-ai/glm-4.6", "openrouter", "glm-4.6"), + ( + "nvidia/nemotron-3-super-120b-a12b", + "openrouter", + "nemotron-3-super", + ), + ("mistralai/devstral-2512", "openrouter", "devstral-2"), + ]; + + RETIRED + .iter() + .map(|(identifier, provider, model)| { + ( + (*identifier).to_string(), + (ProviderId::new(*provider), ModelId::new(*model)), + ) + }) + .collect() +} + +fn build_model_indexes( + models: &[Model], +) -> (OfferingIndex, CanonicalCandidates, AliasCandidates) { + let mut offering_index = HashMap::new(); + let mut canonical_candidates = HashMap::>::new(); + let mut alias_candidates = HashMap::>::new(); for (idx, model) in models.iter().enumerate() { - index.insert(model.id.clone(), idx); + offering_index.insert((model.provider.clone(), model.id.clone()), idx); + canonical_candidates + .entry(model.id.clone()) + .or_default() + .push(idx); for alias in &model.aliases { - index.insert(alias.clone(), idx); + alias_candidates.entry(alias.clone()).or_default().push(idx); } } - index + (offering_index, canonical_candidates, alias_candidates) +} + +fn adopt_legacy_models( + mut overrides: LlmCatalogSettings, + builtins: &LlmCatalogSettings, +) -> Result { + let legacy_models = std::mem::take(&mut overrides.models); + for (identifier, mut settings) in legacy_models { + let provider = match settings.provider.take() { + Some(provider) if !provider.is_empty() => ProviderId::new(provider), + _ => unique_builtin_provider_for_identifier(builtins, &identifier) + .ok_or_else(|| CatalogBuildError::AmbiguousLegacyModelProvider { + model: identifier.clone(), + })?, + }; + let canonical_model = builtins + .providers + .get(provider.as_str()) + .and_then(|provider_settings| { + provider_settings + .models + .iter() + .find(|(model_id, model_settings)| { + model_id.as_str() == identifier + || model_settings + .aliases + .as_ref() + .is_some_and(|aliases| aliases.iter().any(|alias| alias == &identifier)) + }) + .map(|(model_id, _)| model_id.clone()) + }) + .unwrap_or(identifier); + let provider_settings = overrides + .providers + .entry(provider.into_inner()) + .or_default(); + let merged = match provider_settings.models.remove(&canonical_model) { + Some(scoped) => merge_model_settings(settings, scoped), + None => settings, + }; + provider_settings.models.insert(canonical_model, merged); + } + Ok(overrides) +} + +fn unique_builtin_provider_for_identifier( + builtins: &LlmCatalogSettings, + identifier: &str, +) -> Option { + let mut matches = builtins.providers.iter().filter_map(|(provider, settings)| { + settings + .models + .iter() + .any(|(model_id, model)| { + model_id == identifier + || model + .aliases + .as_ref() + .is_some_and(|aliases| aliases.iter().any(|alias| alias == identifier)) + }) + .then(|| ProviderId::new(provider)) + }); + let provider = matches.next()?; + matches.next().is_none().then_some(provider) +} + +fn normalized_model_settings( + settings: &LlmCatalogSettings, +) -> Result, CatalogBuildError> { + let mut normalized = HashMap::new(); + + for (provider, provider_settings) in &settings.providers { + let provider_id = ProviderId::new(provider); + for (model_id, model_settings) in &provider_settings.models { + normalized.insert( + (provider_id.clone(), model_id.clone()), + model_settings.clone(), + ); + } + } + + // Legacy top-level rows remain an input-only compatibility shape. A + // provider is required here; fabro-config performs catalog-aware adoption + // for provider-less rows before constructing these settings. + for (model_id, model_settings) in &settings.models { + let provider = required_model_string(model_id, model_settings.provider.as_ref(), "provider")?; + let key = (ProviderId::new(provider), model_id.clone()); + let merged = match normalized.remove(&key) { + Some(scoped) => merge_model_settings(model_settings.clone(), scoped), + None => model_settings.clone(), + }; + normalized.insert(key, merged); + } + + Ok(normalized) } fn merge_catalog_settings( @@ -1115,9 +1531,24 @@ fn merge_provider_settings( priority: higher.priority.or(fallback.priority), enabled: higher.enabled.or(fallback.enabled), aliases: higher.aliases.or(fallback.aliases), + models: merge_model_maps(higher.models, fallback.models), } } +fn merge_model_maps( + higher: HashMap, + mut fallback: HashMap, +) -> HashMap { + for (id, model) in higher { + let model = match fallback.remove(&id) { + Some(fallback_model) => merge_model_settings(model, fallback_model), + None => model, + }; + fallback.insert(id, model); + } + fallback +} + fn merge_model_settings( higher: ModelCatalogSettings, fallback: ModelCatalogSettings, @@ -1418,7 +1849,7 @@ fn build_model( let speed_costs = build_speed_costs(model_id, settings.costs.as_ref(), &controls)?; let model = Model { - id: model_id.to_string(), + id: ModelId::new(model_id), provider: provider.id.clone(), family, display_name, @@ -1436,6 +1867,12 @@ fn build_model( small_default: settings.small_default.unwrap_or_default(), configured: false, }; + if settings.api_id.as_deref() == Some("") { + return Err(CatalogBuildError::EmptyModelApiId { + provider: provider.id.clone(), + model: ModelId::new(model_id), + }); + } let catalog_settings = CatalogModelSettings { api_id: settings .api_id @@ -1458,7 +1895,7 @@ fn warn_multiple_probe_models(models_with_settings: &[(Model, CatalogModelSettin probes_by_provider .entry(model.provider.clone()) .or_default() - .push(model.id.clone()); + .push(model.id.to_string()); } } @@ -1675,16 +2112,20 @@ fn register_provider_identifier( } fn register_model_identifier( - identifiers: &mut BTreeMap, + identifiers: &mut BTreeMap, identifier: String, - owner: String, + owner: ModelId, + provider: &ProviderId, ) -> Result<(), CatalogBuildError> { match identifiers.get(&identifier) { - Some(existing) if existing != &owner => Err(CatalogBuildError::DuplicateModelIdentifier { - identifier, - first: existing.clone(), - second: owner, - }), + Some(existing) if existing != &owner => { + Err(CatalogBuildError::DuplicateProviderModelIdentifier { + provider: provider.clone(), + identifier, + first: existing.clone(), + second: owner, + }) + } _ => { identifiers.insert(identifier, owner); Ok(()) @@ -1733,6 +2174,20 @@ fn validate_builtin_fragment( }); } } + let provider = fragment + .providers + .get(expected) + .expect("provider count and ID were validated"); + if !fragment.models.is_empty() && !provider.models.is_empty() { + // Embedded fragments are canonical output rather than compatibility + // inputs; mixing shapes would make ownership unclear. + return Err(CatalogBuildError::BuiltinModelProviderMismatch { + path: path.to_string(), + model: "".to_string(), + expected: expected.to_string(), + actual: "top-level models".to_string(), + }); + } Ok(()) } @@ -1762,6 +2217,255 @@ mod tests { toml::from_str(source).expect("fixture should parse as an LLM settings layer") } + const PORTABLE_MODEL_SETTINGS: &str = r#" +[providers.openai] +display_name = "OpenAI" +adapter = "openai" +priority = 90 + +[providers.openai.models."gpt-5.6-sol"] +display_name = "GPT-5.6 Sol" +family = "gpt-5" +aliases = ["gpt-56-sol"] +default = true + +[providers.openai.models."gpt-5.6-sol".limits] +context_window = 1000 + +[providers.openai.models."gpt-5.6-sol".features] +tools = true +vision = false +reasoning = true + +[providers.openrouter] +display_name = "OpenRouter" +adapter = "openai_compatible" +base_url = "https://openrouter.invalid/v1" +priority = 25 + +[providers.openrouter.models."gpt-5.6-sol"] +api_id = "openai/gpt-5.6-sol" +display_name = "GPT-5.6 Sol (via OpenRouter)" +family = "gpt-5" +aliases = ["gpt-56-sol"] +default = true + +[providers.openrouter.models."gpt-5.6-sol".limits] +context_window = 1000 + +[providers.openrouter.models."gpt-5.6-sol".features] +tools = true +vision = false +reasoning = true +"#; + + fn portable_catalog() -> Catalog { + Catalog::from_settings(&minimal_settings(PORTABLE_MODEL_SETTINGS)) + .expect("portable fixture should build") + } + + #[test] + fn provider_aware_catalog_allows_shared_canonical_ids_and_aliases() { + let catalog = portable_catalog(); + let openai = catalog + .model_on_provider(&ProviderId::new("openai"), "gpt-56-sol") + .expect("OpenAI alias should resolve"); + let openrouter = catalog + .model_on_provider(&ProviderId::new("openrouter"), "gpt-56-sol") + .expect("OpenRouter alias should resolve"); + + assert_eq!(openai.id.as_str(), "gpt-5.6-sol"); + assert_eq!(openrouter.id.as_str(), "gpt-5.6-sol"); + assert_ne!(openai.provider, openrouter.provider); + assert_eq!( + catalog + .model_settings_for(openai) + .expect("OpenAI settings should exist") + .api_id, + "gpt-5.6-sol" + ); + assert_eq!( + catalog + .model_settings_for(openrouter) + .expect("OpenRouter settings should exist") + .api_id, + "openai/gpt-5.6-sol" + ); + } + + #[test] + fn provider_aware_selection_uses_eligibility_priority_and_explicit_pin() { + let catalog = portable_catalog(); + let openai = ProviderId::new("openai"); + let openrouter = ProviderId::new("openrouter"); + + assert_eq!( + catalog + .select_model("gpt-56-sol", None, std::slice::from_ref(&openai)) + .unwrap() + .provider, + openai + ); + assert_eq!( + catalog + .select_model("gpt-56-sol", None, std::slice::from_ref(&openrouter)) + .unwrap() + .provider, + openrouter + ); + assert_eq!( + catalog + .select_model("gpt-56-sol", None, &[openrouter.clone(), openai.clone()]) + .unwrap() + .provider, + openai + ); + assert_eq!( + catalog + .select_model( + "gpt-56-sol", + Some(&openrouter), + &[openai.clone(), openrouter.clone()], + ) + .unwrap() + .provider, + openrouter + ); + assert!(matches!( + catalog.select_model("gpt-56-sol", Some(&openrouter), &[openai]), + Err(ModelSelectionError::UnavailableProvider { provider }) + if provider == openrouter + )); + } + + #[test] + fn provider_aware_selection_ties_by_canonical_provider_id() { + let settings = PORTABLE_MODEL_SETTINGS + .replace("priority = 90", "priority = 25"); + let catalog = Catalog::from_settings(&minimal_settings(&settings)).unwrap(); + + assert_eq!( + catalog + .select_model( + "gpt-56-sol", + None, + &[ProviderId::new("openrouter"), ProviderId::new("openai")], + ) + .unwrap() + .provider, + ProviderId::new("openai") + ); + } + + #[test] + fn canonical_id_candidates_shadow_cross_provider_alias_candidates() { + let settings = minimal_settings( + r#" +[providers.canonical] +adapter = "openai" +priority = 1 + +[providers.canonical.models.pin] +display_name = "Canonical Pin" +family = "pin" +default = true +[providers.canonical.models.pin.limits] +context_window = 1000 +[providers.canonical.models.pin.features] +tools = false +vision = false +reasoning = false + +[providers.alias] +adapter = "openai" +priority = 100 + +[providers.alias.models.other] +display_name = "Alias Pin" +family = "pin" +aliases = ["pin"] +default = true +[providers.alias.models.other.limits] +context_window = 1000 +[providers.alias.models.other.features] +tools = false +vision = false +reasoning = false +"#, + ); + let catalog = Catalog::from_settings(&settings).unwrap(); + let canonical = ProviderId::new("canonical"); + let alias = ProviderId::new("alias"); + + assert_eq!( + catalog + .select_model("pin", None, &[alias.clone(), canonical.clone()]) + .unwrap() + .provider, + canonical + ); + assert!(matches!( + catalog.select_model("pin", None, std::slice::from_ref(&alias)), + Err(ModelSelectionError::NoEligibleOffering { selector }) if selector == "pin" + )); + assert_eq!( + catalog + .select_model("pin", Some(&alias), std::slice::from_ref(&alias)) + .unwrap() + .id + .as_str(), + "other" + ); + } + + #[test] + fn same_provider_identifier_collision_is_rejected() { + let source = PORTABLE_MODEL_SETTINGS.replace( + "[providers.openai.models.\"gpt-5.6-sol\".limits]", + r#"[providers.openai.models.other] +display_name = "Other" +family = "gpt-5" +aliases = ["gpt-56-sol"] +[providers.openai.models.other.limits] +context_window = 1000 +[providers.openai.models.other.features] +tools = true +vision = false +reasoning = true + +[providers.openai.models."gpt-5.6-sol".limits]"#, + ); + let err = Catalog::from_settings(&minimal_settings(&source)).unwrap_err(); + + assert!(matches!( + err, + CatalogBuildError::DuplicateProviderModelIdentifier { + provider, + identifier, + first, + second, + } if provider == ProviderId::new("openai") + && identifier == "gpt-56-sol" + && first.as_str() == "gpt-5.6-sol" + && second.as_str() == "other" + )); + } + + #[test] + fn explicitly_empty_api_id_is_rejected() { + let settings = PORTABLE_MODEL_SETTINGS.replace( + "display_name = \"GPT-5.6 Sol\"", + "api_id = \"\"\ndisplay_name = \"GPT-5.6 Sol\"", + ); + let err = Catalog::from_settings(&minimal_settings(&settings)).unwrap_err(); + + assert!(matches!( + err, + CatalogBuildError::EmptyModelApiId { provider, model } + if provider == ProviderId::new("openai") && model.as_str() == "gpt-5.6-sol" + )); + } + const BEDROCK_SIGV4_LAYER: &str = r#" [providers.bedrock] adapter = "bedrock" @@ -2807,8 +3511,15 @@ reasoning = false assert!(matches!( err, - CatalogBuildError::DuplicateModelIdentifier { identifier, first, second } - if identifier == "shared" && first == "one" && second == "two" + CatalogBuildError::DuplicateProviderModelIdentifier { + provider, + identifier, + first, + second, + } if provider == ProviderId::new("test") + && identifier == "shared" + && first.as_str() == "one" + && second.as_str() == "two" )); } diff --git a/lib/crates/fabro-model/src/catalog/providers/anthropic.toml b/lib/crates/fabro-model/src/catalog/providers/anthropic.toml index 0a1587197..95649e214 100644 --- a/lib/crates/fabro-model/src/catalog/providers/anthropic.toml +++ b/lib/crates/fabro-model/src/catalog/providers/anthropic.toml @@ -9,18 +9,16 @@ priority = 100 credentials = ["env:ANTHROPIC_API_KEY", "vault:ANTHROPIC_API_KEY"] header = { custom = "x-api-key" } -[models."claude-fable-5"] -provider = "anthropic" -api_id = "claude-fable-5" +[providers.anthropic.models."claude-fable-5"] display_name = "Claude Fable 5" family = "claude-5" aliases = ["fable", "claude-fable"] -[models."claude-fable-5".limits] +[providers.anthropic.models."claude-fable-5".limits] context_window = 1000000 max_output = 128000 -[models."claude-fable-5".features] +[providers.anthropic.models."claude-fable-5".features] tools = true vision = true reasoning = true @@ -28,14 +26,12 @@ reasoning_effort = "always_adaptive" prompt_cache = true sampling_params = false -[models."claude-fable-5".costs] +[providers.anthropic.models."claude-fable-5".costs] input_cost_per_mtok = 10.0 output_cost_per_mtok = 50.0 cache_input_cost_per_mtok = 1.0 -[models."claude-opus-4-8"] -provider = "anthropic" -api_id = "claude-opus-4-8" +[providers.anthropic.models."claude-opus-4-8"] display_name = "Claude Opus 4.8" family = "claude-4" training = "2026-01-01" @@ -43,11 +39,11 @@ knowledge_cutoff = "Jan 2026" estimated_output_tps = 25 aliases = ["opus", "claude-opus"] -[models."claude-opus-4-8".limits] +[providers.anthropic.models."claude-opus-4-8".limits] context_window = 1000000 max_output = 128000 -[models."claude-opus-4-8".features] +[providers.anthropic.models."claude-opus-4-8".features] tools = true vision = true reasoning = true @@ -55,33 +51,31 @@ reasoning_effort = "levels" prompt_cache = true sampling_params = false -[models."claude-opus-4-8".controls] +[providers.anthropic.models."claude-opus-4-8".controls] speed = ["fast"] -[models."claude-opus-4-8".costs] +[providers.anthropic.models."claude-opus-4-8".costs] input_cost_per_mtok = 5.0 output_cost_per_mtok = 25.0 cache_input_cost_per_mtok = 0.5 -[models."claude-opus-4-8".costs.speed.fast] +[providers.anthropic.models."claude-opus-4-8".costs.speed.fast] input_cost_per_mtok = 10.0 output_cost_per_mtok = 50.0 cache_input_cost_per_mtok = 1.0 -[models."claude-opus-4-7"] -provider = "anthropic" -api_id = "claude-opus-4-7" +[providers.anthropic.models."claude-opus-4-7"] display_name = "Claude Opus 4.7" family = "claude-4" training = "2025-08-01" knowledge_cutoff = "May 2025" estimated_output_tps = 25 -[models."claude-opus-4-7".limits] +[providers.anthropic.models."claude-opus-4-7".limits] context_window = 1000000 max_output = 128000 -[models."claude-opus-4-7".features] +[providers.anthropic.models."claude-opus-4-7".features] tools = true vision = true reasoning = true @@ -89,82 +83,76 @@ reasoning_effort = "levels" prompt_cache = true sampling_params = false -[models."claude-opus-4-7".controls] +[providers.anthropic.models."claude-opus-4-7".controls] speed = ["fast"] -[models."claude-opus-4-7".costs] +[providers.anthropic.models."claude-opus-4-7".costs] input_cost_per_mtok = 5.0 output_cost_per_mtok = 25.0 cache_input_cost_per_mtok = 0.5 -[models."claude-opus-4-7".costs.speed.fast] +[providers.anthropic.models."claude-opus-4-7".costs.speed.fast] input_cost_per_mtok = 30.0 output_cost_per_mtok = 150.0 cache_input_cost_per_mtok = 3.0 -[models."claude-opus-4-6"] -provider = "anthropic" -api_id = "claude-opus-4-6" +[providers.anthropic.models."claude-opus-4-6"] display_name = "Claude Opus 4.6" family = "claude-4" training = "2025-08-01" knowledge_cutoff = "May 2025" estimated_output_tps = 25 -[models."claude-opus-4-6".limits] +[providers.anthropic.models."claude-opus-4-6".limits] context_window = 1000000 max_output = 128000 -[models."claude-opus-4-6".features] +[providers.anthropic.models."claude-opus-4-6".features] tools = true vision = true reasoning = true reasoning_effort = "levels" prompt_cache = true -[models."claude-opus-4-6".controls] +[providers.anthropic.models."claude-opus-4-6".controls] speed = ["fast"] -[models."claude-opus-4-6".costs] +[providers.anthropic.models."claude-opus-4-6".costs] input_cost_per_mtok = 5.0 output_cost_per_mtok = 25.0 cache_input_cost_per_mtok = 0.5 -[models."claude-opus-4-6".costs.speed.fast] +[providers.anthropic.models."claude-opus-4-6".costs.speed.fast] input_cost_per_mtok = 30.0 output_cost_per_mtok = 150.0 cache_input_cost_per_mtok = 3.0 -[models."claude-sonnet-4-5"] -provider = "anthropic" -api_id = "claude-sonnet-4-5" +[providers.anthropic.models."claude-sonnet-4-5"] display_name = "Claude Sonnet 4.5" family = "claude-4" training = "2025-08-01" knowledge_cutoff = "May 2025" estimated_output_tps = 50 -[models."claude-sonnet-4-5".limits] +[providers.anthropic.models."claude-sonnet-4-5".limits] context_window = 200000 max_output = 64000 -[models."claude-sonnet-4-5".features] +[providers.anthropic.models."claude-sonnet-4-5".features] tools = true vision = true reasoning = true prompt_cache = true -[models."claude-sonnet-4-5".controls] +[providers.anthropic.models."claude-sonnet-4-5".controls] reasoning_effort = ["low", "medium", "high", "xhigh", "max"] -[models."claude-sonnet-4-5".costs] +[providers.anthropic.models."claude-sonnet-4-5".costs] input_cost_per_mtok = 3.0 output_cost_per_mtok = 15.0 cache_input_cost_per_mtok = 0.3 -[models."claude-sonnet-4-6"] -provider = "anthropic" -api_id = "claude-sonnet-4-6" +[providers.anthropic.models."claude-sonnet-4-6"] display_name = "Claude Sonnet 4.6" family = "claude-4" training = "2025-08-01" @@ -173,25 +161,23 @@ default = true estimated_output_tps = 50 aliases = ["sonnet", "claude-sonnet"] -[models."claude-sonnet-4-6".limits] +[providers.anthropic.models."claude-sonnet-4-6".limits] context_window = 200000 max_output = 64000 -[models."claude-sonnet-4-6".features] +[providers.anthropic.models."claude-sonnet-4-6".features] tools = true vision = true reasoning = true reasoning_effort = "levels" prompt_cache = true -[models."claude-sonnet-4-6".costs] +[providers.anthropic.models."claude-sonnet-4-6".costs] input_cost_per_mtok = 3.0 output_cost_per_mtok = 15.0 cache_input_cost_per_mtok = 0.3 -[models."claude-haiku-4-5"] -provider = "anthropic" -api_id = "claude-haiku-4-5" +[providers.anthropic.models."claude-haiku-4-5"] display_name = "Claude Haiku 4.5" family = "claude-4" training = "2025-08-01" @@ -201,17 +187,17 @@ aliases = ["haiku", "claude-haiku"] probe = true small_default = true -[models."claude-haiku-4-5".limits] +[providers.anthropic.models."claude-haiku-4-5".limits] context_window = 200000 max_output = 8192 -[models."claude-haiku-4-5".features] +[providers.anthropic.models."claude-haiku-4-5".features] tools = true vision = true reasoning = false prompt_cache = true -[models."claude-haiku-4-5".costs] +[providers.anthropic.models."claude-haiku-4-5".costs] input_cost_per_mtok = 0.8 output_cost_per_mtok = 4.0 cache_input_cost_per_mtok = 0.08 diff --git a/lib/crates/fabro-model/src/catalog/providers/bedrock-openai.toml b/lib/crates/fabro-model/src/catalog/providers/bedrock-openai.toml index 20b011b9c..21384d9b2 100644 --- a/lib/crates/fabro-model/src/catalog/providers/bedrock-openai.toml +++ b/lib/crates/fabro-model/src/catalog/providers/bedrock-openai.toml @@ -35,41 +35,41 @@ credentials = [ # [llm.providers.bedrock-openai] # enabled = true -[models."openai.gpt-5.5"] -provider = "bedrock-openai" +[providers.bedrock-openai.models."gpt-5.5"] +api_id = "openai.gpt-5.5" display_name = "GPT-5.5 (Bedrock)" family = "gpt-5" default = true -[models."openai.gpt-5.5".limits] +[providers.bedrock-openai.models."gpt-5.5".limits] context_window = 272000 max_output = 128000 -[models."openai.gpt-5.5".features] +[providers.bedrock-openai.models."gpt-5.5".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."openai.gpt-5.5".costs] +[providers.bedrock-openai.models."gpt-5.5".costs] input_cost_per_mtok = 5.5 output_cost_per_mtok = 33.0 -[models."openai.gpt-5.4"] -provider = "bedrock-openai" +[providers.bedrock-openai.models."gpt-5.4"] +api_id = "openai.gpt-5.4" display_name = "GPT-5.4 (Bedrock)" family = "gpt-5" -[models."openai.gpt-5.4".limits] +[providers.bedrock-openai.models."gpt-5.4".limits] context_window = 272000 max_output = 128000 -[models."openai.gpt-5.4".features] +[providers.bedrock-openai.models."gpt-5.4".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."openai.gpt-5.4".costs] +[providers.bedrock-openai.models."gpt-5.4".costs] input_cost_per_mtok = 2.75 output_cost_per_mtok = 16.5 diff --git a/lib/crates/fabro-model/src/catalog/providers/bedrock.toml b/lib/crates/fabro-model/src/catalog/providers/bedrock.toml index f6f217a4f..aeed22e46 100644 --- a/lib/crates/fabro-model/src/catalog/providers/bedrock.toml +++ b/lib/crates/fabro-model/src/catalog/providers/bedrock.toml @@ -45,68 +45,67 @@ credentials = [ # file because its Bedrock deployment pins sampling parameters and requires an # extra data-sharing opt-in. -[models."us.anthropic.claude-sonnet-4-6"] -provider = "bedrock" +[providers.bedrock.models."claude-sonnet-4-6"] +api_id = "us.anthropic.claude-sonnet-4-6" display_name = "Claude Sonnet 4.6 (Bedrock)" family = "claude-4" billing_policy = "anthropic" default = true -[models."us.anthropic.claude-sonnet-4-6".limits] +[providers.bedrock.models."claude-sonnet-4-6".limits] context_window = 1000000 max_output = 64000 -[models."us.anthropic.claude-sonnet-4-6".features] +[providers.bedrock.models."claude-sonnet-4-6".features] tools = true vision = true reasoning = true prompt_cache = true -[models."us.anthropic.claude-sonnet-4-6".costs] +[providers.bedrock.models."claude-sonnet-4-6".costs] input_cost_per_mtok = 3.0 output_cost_per_mtok = 15.0 cache_input_cost_per_mtok = 0.3 -[models."us.anthropic.claude-opus-4-8"] -provider = "bedrock" +[providers.bedrock.models."claude-opus-4-8"] +api_id = "us.anthropic.claude-opus-4-8" display_name = "Claude Opus 4.8 (Bedrock)" family = "claude-4" billing_policy = "anthropic" -[models."us.anthropic.claude-opus-4-8".limits] +[providers.bedrock.models."claude-opus-4-8".limits] context_window = 1000000 max_output = 128000 -[models."us.anthropic.claude-opus-4-8".features] +[providers.bedrock.models."claude-opus-4-8".features] tools = true vision = true reasoning = true prompt_cache = true -[models."us.anthropic.claude-opus-4-8".costs] +[providers.bedrock.models."claude-opus-4-8".costs] input_cost_per_mtok = 5.0 output_cost_per_mtok = 25.0 cache_input_cost_per_mtok = 0.5 -[models."us.anthropic.claude-haiku-4-5"] -provider = "bedrock" +[providers.bedrock.models."claude-haiku-4-5"] api_id = "us.anthropic.claude-haiku-4-5-20251001-v1:0" display_name = "Claude Haiku 4.5 (Bedrock)" family = "claude-4" billing_policy = "anthropic" small_default = true -[models."us.anthropic.claude-haiku-4-5".limits] +[providers.bedrock.models."claude-haiku-4-5".limits] context_window = 200000 max_output = 64000 -[models."us.anthropic.claude-haiku-4-5".features] +[providers.bedrock.models."claude-haiku-4-5".features] tools = true vision = true reasoning = false prompt_cache = true -[models."us.anthropic.claude-haiku-4-5".costs] +[providers.bedrock.models."claude-haiku-4-5".costs] input_cost_per_mtok = 1.0 output_cost_per_mtok = 5.0 cache_input_cost_per_mtok = 0.1 @@ -116,149 +115,142 @@ cache_input_cost_per_mtok = 0.1 # GPT-5.5/5.4 are NOT here: on Bedrock they are Responses-API-only on the # bedrock-mantle endpoint (no Converse), a named follow-up route. -[models."openai.gpt-oss-120b"] -provider = "bedrock" +[providers.bedrock.models."gpt-oss-120b"] api_id = "openai.gpt-oss-120b-1:0" display_name = "GPT-OSS 120B (Bedrock)" family = "gpt-oss" billing_policy = "openai" agent_profile = "openai" -[models."openai.gpt-oss-120b".limits] +[providers.bedrock.models."gpt-oss-120b".limits] context_window = 128000 max_output = 16384 -[models."openai.gpt-oss-120b".features] +[providers.bedrock.models."gpt-oss-120b".features] tools = true vision = false reasoning = true -[models."openai.gpt-oss-120b".costs] +[providers.bedrock.models."gpt-oss-120b".costs] input_cost_per_mtok = 0.15 output_cost_per_mtok = 0.60 -[models."openai.gpt-oss-20b"] -provider = "bedrock" +[providers.bedrock.models."gpt-oss-20b"] api_id = "openai.gpt-oss-20b-1:0" display_name = "GPT-OSS 20B (Bedrock)" family = "gpt-oss" billing_policy = "openai" agent_profile = "openai" -[models."openai.gpt-oss-20b".limits] +[providers.bedrock.models."gpt-oss-20b".limits] context_window = 128000 max_output = 16384 -[models."openai.gpt-oss-20b".features] +[providers.bedrock.models."gpt-oss-20b".features] tools = true vision = false reasoning = true -[models."openai.gpt-oss-20b".costs] +[providers.bedrock.models."gpt-oss-20b".costs] input_cost_per_mtok = 0.07 output_cost_per_mtok = 0.30 # ---------- Amazon Nova ---------- -[models."amazon.nova-2-lite"] -provider = "bedrock" +[providers.bedrock.models."nova-2-lite"] api_id = "global.amazon.nova-2-lite-v1:0" display_name = "Nova 2 Lite (Bedrock)" family = "nova-2" billing_policy = "openai" agent_profile = "openai" -[models."amazon.nova-2-lite".limits] +[providers.bedrock.models."nova-2-lite".limits] context_window = 1000000 # Bedrock caps Nova output at 65535 (2^16 - 1); 65536 trips # "maximum tokens exceeds the model limit of 65535" since the prompt handler # defaults max_tokens to max_output. max_output = 65535 -[models."amazon.nova-2-lite".features] +[providers.bedrock.models."nova-2-lite".features] tools = true vision = true reasoning = false -[models."amazon.nova-2-lite".costs] +[providers.bedrock.models."nova-2-lite".costs] input_cost_per_mtok = 0.30 output_cost_per_mtok = 2.50 # ---------- Open-weights ---------- -[models."meta.llama4-maverick"] -provider = "bedrock" +[providers.bedrock.models."llama-4-maverick"] api_id = "us.meta.llama4-maverick-17b-instruct-v1:0" display_name = "Llama 4 Maverick (Bedrock)" family = "llama-4" billing_policy = "openai" agent_profile = "openai" -[models."meta.llama4-maverick".limits] +[providers.bedrock.models."llama-4-maverick".limits] context_window = 1000000 max_output = 8192 -[models."meta.llama4-maverick".features] +[providers.bedrock.models."llama-4-maverick".features] tools = true vision = true reasoning = false -[models."mistral.mistral-large-3"] -provider = "bedrock" +[providers.bedrock.models."mistral-large-3"] api_id = "mistral.mistral-large-3-675b-instruct" display_name = "Mistral Large 3 (Bedrock)" family = "mistral-large" billing_policy = "openai" agent_profile = "openai" -[models."mistral.mistral-large-3".limits] +[providers.bedrock.models."mistral-large-3".limits] context_window = 256000 max_output = 32768 -[models."mistral.mistral-large-3".features] +[providers.bedrock.models."mistral-large-3".features] tools = true vision = true reasoning = false -[models."mistral.mistral-large-3".costs] +[providers.bedrock.models."mistral-large-3".costs] input_cost_per_mtok = 0.50 output_cost_per_mtok = 1.50 -[models."mistral.devstral-2"] -provider = "bedrock" +[providers.bedrock.models."devstral-2"] api_id = "mistral.devstral-2-123b" display_name = "Devstral 2 (Bedrock)" family = "devstral" billing_policy = "openai" agent_profile = "openai" -[models."mistral.devstral-2".limits] +[providers.bedrock.models."devstral-2".limits] context_window = 256000 max_output = 32768 -[models."mistral.devstral-2".features] +[providers.bedrock.models."devstral-2".features] tools = true vision = false reasoning = false -[models."deepseek.v3-2"] -provider = "bedrock" +[providers.bedrock.models."deepseek-v3.2"] api_id = "deepseek.v3.2" display_name = "DeepSeek V3.2 (Bedrock)" family = "deepseek-v3" billing_policy = "openai" agent_profile = "openai" -[models."deepseek.v3-2".limits] +[providers.bedrock.models."deepseek-v3.2".limits] context_window = 164000 max_output = 8192 -[models."deepseek.v3-2".features] +[providers.bedrock.models."deepseek-v3.2".features] tools = true vision = false reasoning = true -[models."deepseek.v3-2".costs] +[providers.bedrock.models."deepseek-v3.2".costs] input_cost_per_mtok = 0.62 output_cost_per_mtok = 1.85 @@ -267,92 +259,90 @@ output_cost_per_mtok = 1.85 # "The provided model identifier is invalid"), so this row needs an explicit # `api_id` confirmed against `aws bedrock list-inference-profiles` before it # ships. Re-add with: -# [models."qwen.qwen3-coder-next"] -# provider = "bedrock" +# [providers.bedrock.models."qwen3-coder-next"] # api_id = "" # display_name = "Qwen3 Coder Next (Bedrock)" # family = "qwen3" # billing_policy = "openai" # agent_profile = "openai" -# [models."qwen.qwen3-coder-next".limits] +# [providers.bedrock.models."qwen3-coder-next".limits] # context_window = 256000 # max_output = 16384 -# [models."qwen.qwen3-coder-next".features] +# [providers.bedrock.models."qwen3-coder-next".features] # tools = true -[models."moonshotai.kimi-k2.5"] -provider = "bedrock" +[providers.bedrock.models."kimi-k2.5"] +api_id = "moonshotai.kimi-k2.5" display_name = "Kimi K2.5 (Bedrock)" family = "kimi-k2" billing_policy = "openai" agent_profile = "openai" -[models."moonshotai.kimi-k2.5".limits] +[providers.bedrock.models."kimi-k2.5".limits] context_window = 262144 max_output = 16384 -[models."moonshotai.kimi-k2.5".features] +[providers.bedrock.models."kimi-k2.5".features] tools = true vision = true reasoning = false -[models."moonshotai.kimi-k2.5".costs] +[providers.bedrock.models."kimi-k2.5".costs] input_cost_per_mtok = 0.60 output_cost_per_mtok = 3.00 -[models."zai.glm-5"] -provider = "bedrock" +[providers.bedrock.models."glm-5"] +api_id = "zai.glm-5" display_name = "GLM 5 (Bedrock)" family = "glm" billing_policy = "openai" agent_profile = "openai" -[models."zai.glm-5".limits] +[providers.bedrock.models."glm-5".limits] context_window = 200000 max_output = 128000 -[models."zai.glm-5".features] +[providers.bedrock.models."glm-5".features] tools = true vision = false reasoning = false -[models."zai.glm-5".costs] +[providers.bedrock.models."glm-5".costs] input_cost_per_mtok = 1.00 output_cost_per_mtok = 3.20 -[models."minimax.minimax-m2.5"] -provider = "bedrock" +[providers.bedrock.models."minimax-m2.5"] +api_id = "minimax.minimax-m2.5" display_name = "MiniMax M2.5 (Bedrock)" family = "minimax-m2" billing_policy = "openai" agent_profile = "openai" -[models."minimax.minimax-m2.5".limits] +[providers.bedrock.models."minimax-m2.5".limits] context_window = 196000 max_output = 8192 -[models."minimax.minimax-m2.5".features] +[providers.bedrock.models."minimax-m2.5".features] tools = true vision = false reasoning = false -[models."minimax.minimax-m2.5".costs] +[providers.bedrock.models."minimax-m2.5".costs] input_cost_per_mtok = 0.30 output_cost_per_mtok = 1.20 -[models."nvidia.nemotron-3-super"] -provider = "bedrock" +[providers.bedrock.models."nemotron-3-super"] api_id = "nvidia.nemotron-super-3-120b" display_name = "Nemotron 3 Super (Bedrock)" family = "nemotron-3" billing_policy = "openai" agent_profile = "openai" -[models."nvidia.nemotron-3-super".limits] +[providers.bedrock.models."nemotron-3-super".limits] context_window = 256000 max_output = 32768 -[models."nvidia.nemotron-3-super".features] +[providers.bedrock.models."nemotron-3-super".features] tools = true vision = false reasoning = false @@ -365,24 +355,24 @@ reasoning = false # reasoning_effort stays undeclared here (requests carrying one are # rejected up front rather than silently dropped). -[models."us.anthropic.claude-fable-5"] -provider = "bedrock" +[providers.bedrock.models."claude-fable-5"] +api_id = "us.anthropic.claude-fable-5" display_name = "Claude Fable 5 (Bedrock)" family = "claude-5" billing_policy = "anthropic" -[models."us.anthropic.claude-fable-5".limits] +[providers.bedrock.models."claude-fable-5".limits] context_window = 1000000 max_output = 128000 -[models."us.anthropic.claude-fable-5".features] +[providers.bedrock.models."claude-fable-5".features] tools = true vision = true reasoning = true prompt_cache = true sampling_params = false -[models."us.anthropic.claude-fable-5".costs] +[providers.bedrock.models."claude-fable-5".costs] input_cost_per_mtok = 10.0 output_cost_per_mtok = 50.0 cache_input_cost_per_mtok = 1.0 diff --git a/lib/crates/fabro-model/src/catalog/providers/gemini.toml b/lib/crates/fabro-model/src/catalog/providers/gemini.toml index 74ffc91de..a03c2a249 100644 --- a/lib/crates/fabro-model/src/catalog/providers/gemini.toml +++ b/lib/crates/fabro-model/src/catalog/providers/gemini.toml @@ -9,9 +9,7 @@ priority = 80 credentials = ["env:GEMINI_API_KEY", "env:GOOGLE_API_KEY", "vault:GEMINI_API_KEY"] header = { custom = "x-goog-api-key" } -[models."gemini-3.1-pro-preview"] -provider = "gemini" -api_id = "gemini-3.1-pro-preview" +[providers.gemini.models."gemini-3.1-pro-preview"] display_name = "Gemini 3.1 Pro (Preview)" family = "gemini-3" training = "2025-01-01" @@ -19,24 +17,22 @@ knowledge_cutoff = "January 2025" estimated_output_tps = 85 aliases = ["gemini-pro"] -[models."gemini-3.1-pro-preview".limits] +[providers.gemini.models."gemini-3.1-pro-preview".limits] context_window = 1048576 max_output = 65536 -[models."gemini-3.1-pro-preview".features] +[providers.gemini.models."gemini-3.1-pro-preview".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."gemini-3.1-pro-preview".costs] +[providers.gemini.models."gemini-3.1-pro-preview".costs] input_cost_per_mtok = 2.0 output_cost_per_mtok = 12.0 cache_input_cost_per_mtok = 0.5 -[models."gemini-3.1-pro-preview-customtools"] -provider = "gemini" -api_id = "gemini-3.1-pro-preview-customtools" +[providers.gemini.models."gemini-3.1-pro-preview-customtools"] display_name = "Gemini 3.1 Pro Custom Tools (Preview)" family = "gemini-3" training = "2025-01-01" @@ -44,24 +40,22 @@ knowledge_cutoff = "January 2025" estimated_output_tps = 85 aliases = ["gemini-customtools"] -[models."gemini-3.1-pro-preview-customtools".limits] +[providers.gemini.models."gemini-3.1-pro-preview-customtools".limits] context_window = 1048576 max_output = 65536 -[models."gemini-3.1-pro-preview-customtools".features] +[providers.gemini.models."gemini-3.1-pro-preview-customtools".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."gemini-3.1-pro-preview-customtools".costs] +[providers.gemini.models."gemini-3.1-pro-preview-customtools".costs] input_cost_per_mtok = 2.0 output_cost_per_mtok = 12.0 cache_input_cost_per_mtok = 0.5 -[models."gemini-3.5-flash"] -provider = "gemini" -api_id = "gemini-3.5-flash" +[providers.gemini.models."gemini-3.5-flash"] display_name = "Gemini 3.5 Flash" family = "gemini-3" training = "2025-01-01" @@ -70,24 +64,22 @@ default = true estimated_output_tps = 150 aliases = ["gemini-35-flash"] -[models."gemini-3.5-flash".limits] +[providers.gemini.models."gemini-3.5-flash".limits] context_window = 1048576 max_output = 65536 -[models."gemini-3.5-flash".features] +[providers.gemini.models."gemini-3.5-flash".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."gemini-3.5-flash".costs] +[providers.gemini.models."gemini-3.5-flash".costs] input_cost_per_mtok = 1.5 output_cost_per_mtok = 9.0 cache_input_cost_per_mtok = 0.15 -[models."gemini-3-flash-preview"] -provider = "gemini" -api_id = "gemini-3-flash-preview" +[providers.gemini.models."gemini-3-flash-preview"] display_name = "Gemini 3 Flash (Preview)" family = "gemini-3" training = "2025-01-01" @@ -95,24 +87,22 @@ knowledge_cutoff = "January 2025" estimated_output_tps = 150 aliases = ["gemini-flash"] -[models."gemini-3-flash-preview".limits] +[providers.gemini.models."gemini-3-flash-preview".limits] context_window = 1048576 max_output = 65536 -[models."gemini-3-flash-preview".features] +[providers.gemini.models."gemini-3-flash-preview".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."gemini-3-flash-preview".costs] +[providers.gemini.models."gemini-3-flash-preview".costs] input_cost_per_mtok = 0.5 output_cost_per_mtok = 3.0 cache_input_cost_per_mtok = 0.125 -[models."gemini-3.1-flash-lite"] -provider = "gemini" -api_id = "gemini-3.1-flash-lite" +[providers.gemini.models."gemini-3.1-flash-lite"] display_name = "Gemini 3.1 Flash Lite" family = "gemini-3" training = "2025-01-01" @@ -121,17 +111,17 @@ estimated_output_tps = 200 aliases = ["gemini-flash-lite", "gemini-3.1-flash-lite-preview"] small_default = true -[models."gemini-3.1-flash-lite".limits] +[providers.gemini.models."gemini-3.1-flash-lite".limits] context_window = 1048576 max_output = 65536 -[models."gemini-3.1-flash-lite".features] +[providers.gemini.models."gemini-3.1-flash-lite".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."gemini-3.1-flash-lite".costs] +[providers.gemini.models."gemini-3.1-flash-lite".costs] input_cost_per_mtok = 0.25 output_cost_per_mtok = 1.5 cache_input_cost_per_mtok = 0.025 diff --git a/lib/crates/fabro-model/src/catalog/providers/inception.toml b/lib/crates/fabro-model/src/catalog/providers/inception.toml index 1d2b08a64..965120f27 100644 --- a/lib/crates/fabro-model/src/catalog/providers/inception.toml +++ b/lib/crates/fabro-model/src/catalog/providers/inception.toml @@ -8,25 +8,23 @@ priority = 40 [providers.inception.auth] credentials = ["env:INCEPTION_API_KEY", "vault:INCEPTION_API_KEY"] -[models."mercury-2"] -provider = "inception" -api_id = "mercury-2" +[providers.inception.models."mercury-2"] display_name = "Mercury 2" family = "mercury" default = true estimated_output_tps = 1000 aliases = ["mercury"] -[models."mercury-2".limits] +[providers.inception.models."mercury-2".limits] context_window = 131072 max_output = 50000 -[models."mercury-2".features] +[providers.inception.models."mercury-2".features] tools = true vision = false reasoning = true reasoning_effort = "levels" -[models."mercury-2".costs] +[providers.inception.models."mercury-2".costs] input_cost_per_mtok = 0.25 output_cost_per_mtok = 0.75 diff --git a/lib/crates/fabro-model/src/catalog/providers/kimi.toml b/lib/crates/fabro-model/src/catalog/providers/kimi.toml index c57779116..daa4b20c2 100644 --- a/lib/crates/fabro-model/src/catalog/providers/kimi.toml +++ b/lib/crates/fabro-model/src/catalog/providers/kimi.toml @@ -8,46 +8,42 @@ priority = 70 [providers.kimi.auth] credentials = ["env:KIMI_API_KEY", "vault:KIMI_API_KEY"] -[models."kimi-k2.5"] -provider = "kimi" -api_id = "kimi-k2.5" +[providers.kimi.models."kimi-k2.5"] display_name = "Kimi K2.5" family = "kimi-k2" training = "2025-10-01" knowledge_cutoff = "October 2025" estimated_output_tps = 50 -[models."kimi-k2.5".limits] +[providers.kimi.models."kimi-k2.5".limits] context_window = 262144 max_output = 32768 -[models."kimi-k2.5".features] +[providers.kimi.models."kimi-k2.5".features] tools = true vision = true reasoning = true prompt_cache = true sampling_params = false -[models."kimi-k2.5".costs] +[providers.kimi.models."kimi-k2.5".costs] input_cost_per_mtok = 0.6 output_cost_per_mtok = 3.0 cache_input_cost_per_mtok = 0.1 -[models."kimi-k3"] -provider = "kimi" -api_id = "kimi-k3" +[providers.kimi.models."kimi-k3"] display_name = "Kimi K3" family = "kimi-k3" default = true aliases = ["kimi"] -[models."kimi-k3".limits] +[providers.kimi.models."kimi-k3".limits] context_window = 1048576 # K3 accepts explicit completion budgets up to 1048576, but Fabro also uses # max_output as the default request budget. Match Kimi's 131072-token default. max_output = 131072 -[models."kimi-k3".features] +[providers.kimi.models."kimi-k3".features] tools = true vision = true reasoning = true @@ -55,10 +51,10 @@ reasoning_effort = "always_adaptive" prompt_cache = true sampling_params = false -[models."kimi-k3".controls] +[providers.kimi.models."kimi-k3".controls] reasoning_effort = ["low", "high", "max"] -[models."kimi-k3".costs] +[providers.kimi.models."kimi-k3".costs] input_cost_per_mtok = 3.0 output_cost_per_mtok = 15.0 cache_input_cost_per_mtok = 0.3 diff --git a/lib/crates/fabro-model/src/catalog/providers/litellm.toml b/lib/crates/fabro-model/src/catalog/providers/litellm.toml index 55f5aef15..1307378c1 100644 --- a/lib/crates/fabro-model/src/catalog/providers/litellm.toml +++ b/lib/crates/fabro-model/src/catalog/providers/litellm.toml @@ -14,18 +14,17 @@ credentials = ["env:LITELLM_API_KEY", "vault:LITELLM_API_KEY"] # enabled = true # base_url = "http://localhost:4000/v1" # -# [llm.models."litellm-gpt-5"] -# provider = "litellm" +# [llm.providers.litellm.models."litellm-gpt-5"] # api_id = "gpt-5" # display_name = "LiteLLM GPT-5" # family = "litellm" # default = true # -# [llm.models."litellm-gpt-5".limits] +# [llm.providers.litellm.models."litellm-gpt-5".limits] # context_window = 128000 # max_output = 8192 # -# [llm.models."litellm-gpt-5".features] +# [llm.providers.litellm.models."litellm-gpt-5".features] # tools = true # vision = false # reasoning = false diff --git a/lib/crates/fabro-model/src/catalog/providers/minimax.toml b/lib/crates/fabro-model/src/catalog/providers/minimax.toml index 172a6fd6a..e68dfc290 100644 --- a/lib/crates/fabro-model/src/catalog/providers/minimax.toml +++ b/lib/crates/fabro-model/src/catalog/providers/minimax.toml @@ -8,24 +8,22 @@ priority = 50 [providers.minimax.auth] credentials = ["env:MINIMAX_API_KEY", "vault:MINIMAX_API_KEY"] -[models."minimax-m2.5"] -provider = "minimax" -api_id = "minimax-m2.5" +[providers.minimax.models."minimax-m2.5"] display_name = "Minimax M2.5" family = "minimax-m2" default = true estimated_output_tps = 45 aliases = ["minimax"] -[models."minimax-m2.5".limits] +[providers.minimax.models."minimax-m2.5".limits] context_window = 196608 max_output = 16384 -[models."minimax-m2.5".features] +[providers.minimax.models."minimax-m2.5".features] tools = true vision = false reasoning = false -[models."minimax-m2.5".costs] +[providers.minimax.models."minimax-m2.5".costs] input_cost_per_mtok = 0.3 output_cost_per_mtok = 1.2 diff --git a/lib/crates/fabro-model/src/catalog/providers/ollama.toml b/lib/crates/fabro-model/src/catalog/providers/ollama.toml index bf3b16f5e..78dc5db69 100644 --- a/lib/crates/fabro-model/src/catalog/providers/ollama.toml +++ b/lib/crates/fabro-model/src/catalog/providers/ollama.toml @@ -9,18 +9,17 @@ enabled = false # Example model. Uncomment after `ollama pull qwen3.5` (and `enabled = true` # above) to expose it through the OpenAI-compatible adapter. # -# [models."qwen3.5"] -# provider = "ollama" +# [providers.ollama.models."qwen3.5"] # api_id = "qwen3.5:latest" # display_name = "Qwen3.5" # family = "qwen3.5" # default = true # aliases = ["ollama-qwen3.5"] # -# [models."qwen3.5".limits] +# [providers.ollama.models."qwen3.5".limits] # context_window = 32768 # -# [models."qwen3.5".features] +# [providers.ollama.models."qwen3.5".features] # tools = true # vision = false # reasoning = false diff --git a/lib/crates/fabro-model/src/catalog/providers/openai.toml b/lib/crates/fabro-model/src/catalog/providers/openai.toml index 56fe7a03c..9ae21c91f 100644 --- a/lib/crates/fabro-model/src/catalog/providers/openai.toml +++ b/lib/crates/fabro-model/src/catalog/providers/openai.toml @@ -8,9 +8,7 @@ priority = 90 [providers.openai.auth] credentials = ["env:OPENAI_API_KEY", "vault:OPENAI_API_KEY", "vault:OPENAI_CODEX"] -[models."gpt-5.6-sol"] -provider = "openai" -api_id = "gpt-5.6-sol" +[providers.openai.models."gpt-5.6-sol"] display_name = "GPT-5.6 Sol" family = "gpt-5" training = "2026-02-16" @@ -18,75 +16,69 @@ knowledge_cutoff = "February 16, 2026" default = true aliases = ["gpt56-sol", "gpt-56-sol", "gpt-5.6", "gpt56", "gpt-56"] -[models."gpt-5.6-sol".limits] +[providers.openai.models."gpt-5.6-sol".limits] context_window = 272000 max_output = 128000 -[models."gpt-5.6-sol".features] +[providers.openai.models."gpt-5.6-sol".features] tools = true vision = true reasoning = true reasoning_effort = "levels" prompt_cache = true -[models."gpt-5.6-sol".costs] +[providers.openai.models."gpt-5.6-sol".costs] input_cost_per_mtok = 5.0 output_cost_per_mtok = 30.0 cache_input_cost_per_mtok = 0.5 -[models."gpt-5.6-terra"] -provider = "openai" -api_id = "gpt-5.6-terra" +[providers.openai.models."gpt-5.6-terra"] display_name = "GPT-5.6 Terra" family = "gpt-5" training = "2026-02-16" knowledge_cutoff = "February 16, 2026" aliases = ["gpt56-terra", "gpt-56-terra"] -[models."gpt-5.6-terra".limits] +[providers.openai.models."gpt-5.6-terra".limits] context_window = 272000 max_output = 128000 -[models."gpt-5.6-terra".features] +[providers.openai.models."gpt-5.6-terra".features] tools = true vision = true reasoning = true reasoning_effort = "levels" prompt_cache = true -[models."gpt-5.6-terra".costs] +[providers.openai.models."gpt-5.6-terra".costs] input_cost_per_mtok = 2.5 output_cost_per_mtok = 15.0 cache_input_cost_per_mtok = 0.25 -[models."gpt-5.6-luna"] -provider = "openai" -api_id = "gpt-5.6-luna" +[providers.openai.models."gpt-5.6-luna"] display_name = "GPT-5.6 Luna" family = "gpt-5" training = "2026-02-16" knowledge_cutoff = "February 16, 2026" aliases = ["gpt56-luna", "gpt-56-luna"] -[models."gpt-5.6-luna".limits] +[providers.openai.models."gpt-5.6-luna".limits] context_window = 272000 max_output = 128000 -[models."gpt-5.6-luna".features] +[providers.openai.models."gpt-5.6-luna".features] tools = true vision = true reasoning = true reasoning_effort = "levels" prompt_cache = true -[models."gpt-5.6-luna".costs] +[providers.openai.models."gpt-5.6-luna".costs] input_cost_per_mtok = 1.0 output_cost_per_mtok = 6.0 cache_input_cost_per_mtok = 0.1 -[models."gpt-5.4"] -provider = "openai" -api_id = "gpt-5.4" +[providers.openai.models."gpt-5.4"] display_name = "GPT-5.4" family = "gpt-5" training = "2025-08-31" @@ -94,24 +86,22 @@ knowledge_cutoff = "April 2025" estimated_output_tps = 70 aliases = ["gpt54", "gpt-54", "gpt-5.2", "gpt5", "gpt-5.3-codex", "codex"] -[models."gpt-5.4".limits] +[providers.openai.models."gpt-5.4".limits] context_window = 272000 max_output = 128000 -[models."gpt-5.4".features] +[providers.openai.models."gpt-5.4".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."gpt-5.4".costs] +[providers.openai.models."gpt-5.4".costs] input_cost_per_mtok = 2.5 output_cost_per_mtok = 15.0 cache_input_cost_per_mtok = 0.25 -[models."gpt-5.5"] -provider = "openai" -api_id = "gpt-5.5" +[providers.openai.models."gpt-5.5"] display_name = "GPT-5.5" family = "gpt-5" training = "2025-12-01" @@ -119,24 +109,22 @@ knowledge_cutoff = "December 2025" estimated_output_tps = 70 aliases = ["gpt55", "gpt-55"] -[models."gpt-5.5".limits] +[providers.openai.models."gpt-5.5".limits] context_window = 272000 max_output = 128000 -[models."gpt-5.5".features] +[providers.openai.models."gpt-5.5".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."gpt-5.5".costs] +[providers.openai.models."gpt-5.5".costs] input_cost_per_mtok = 5.0 output_cost_per_mtok = 30.0 cache_input_cost_per_mtok = 0.5 -[models."gpt-5.5-pro"] -provider = "openai" -api_id = "gpt-5.5-pro" +[providers.openai.models."gpt-5.5-pro"] display_name = "GPT-5.5 Pro" family = "gpt-5" training = "2025-12-01" @@ -144,24 +132,22 @@ knowledge_cutoff = "December 2025" estimated_output_tps = 20 aliases = ["gpt55-pro", "gpt-55-pro"] -[models."gpt-5.5-pro".limits] +[providers.openai.models."gpt-5.5-pro".limits] context_window = 1050000 max_output = 128000 -[models."gpt-5.5-pro".features] +[providers.openai.models."gpt-5.5-pro".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."gpt-5.5-pro".costs] +[providers.openai.models."gpt-5.5-pro".costs] input_cost_per_mtok = 30.0 output_cost_per_mtok = 180.0 cache_input_cost_per_mtok = 3.0 -[models."gpt-5.4-pro"] -provider = "openai" -api_id = "gpt-5.4-pro" +[providers.openai.models."gpt-5.4-pro"] display_name = "GPT-5.4 Pro" family = "gpt-5" training = "2025-08-31" @@ -169,24 +155,22 @@ knowledge_cutoff = "April 2025" estimated_output_tps = 20 aliases = ["gpt54-pro", "gpt-54-pro"] -[models."gpt-5.4-pro".limits] +[providers.openai.models."gpt-5.4-pro".limits] context_window = 1047576 max_output = 128000 -[models."gpt-5.4-pro".features] +[providers.openai.models."gpt-5.4-pro".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."gpt-5.4-pro".costs] +[providers.openai.models."gpt-5.4-pro".costs] input_cost_per_mtok = 30.0 output_cost_per_mtok = 180.0 cache_input_cost_per_mtok = 3.0 -[models."gpt-5.4-mini"] -provider = "openai" -api_id = "gpt-5.4-mini" +[providers.openai.models."gpt-5.4-mini"] display_name = "GPT-5.4 Mini" family = "gpt-5" training = "2025-08-31" @@ -196,17 +180,17 @@ aliases = ["gpt54-mini", "gpt-54-mini", "gpt-5.3-codex-spark", "codex-spark"] probe = true small_default = true -[models."gpt-5.4-mini".limits] +[providers.openai.models."gpt-5.4-mini".limits] context_window = 272000 max_output = 128000 -[models."gpt-5.4-mini".features] +[providers.openai.models."gpt-5.4-mini".features] tools = true vision = true reasoning = true reasoning_effort = "levels" -[models."gpt-5.4-mini".costs] +[providers.openai.models."gpt-5.4-mini".costs] input_cost_per_mtok = 0.75 output_cost_per_mtok = 4.5 cache_input_cost_per_mtok = 0.075 diff --git a/lib/crates/fabro-model/src/catalog/providers/openrouter.toml b/lib/crates/fabro-model/src/catalog/providers/openrouter.toml index 86fd43319..bad07290a 100644 --- a/lib/crates/fabro-model/src/catalog/providers/openrouter.toml +++ b/lib/crates/fabro-model/src/catalog/providers/openrouter.toml @@ -33,262 +33,249 @@ credentials = ["env:OPENROUTER_API_KEY", "vault:OPENROUTER_API_KEY"] # best-effort estimates; OpenRouter returns the authoritative usage.cost # in-band on every response. -[models."anthropic/claude-opus-4-7"] -provider = "openrouter" +[providers.openrouter.models."claude-opus-4-7"] api_id = "anthropic/claude-opus-4.7" display_name = "Claude Opus 4.7 (via OpenRouter)" family = "claude-4" billing_policy = "anthropic" -[models."anthropic/claude-opus-4-7".limits] +[providers.openrouter.models."claude-opus-4-7".limits] context_window = 1000000 max_output = 128000 -[models."anthropic/claude-opus-4-7".features] +[providers.openrouter.models."claude-opus-4-7".features] tools = true vision = true reasoning = true prompt_cache = true -[models."anthropic/claude-opus-4-7".costs] +[providers.openrouter.models."claude-opus-4-7".costs] input_cost_per_mtok = 5.0 output_cost_per_mtok = 25.0 cache_input_cost_per_mtok = 0.5 -[models."anthropic/claude-sonnet-4-6"] -provider = "openrouter" +[providers.openrouter.models."claude-sonnet-4-6"] api_id = "anthropic/claude-sonnet-4.6" display_name = "Claude Sonnet 4.6 (via OpenRouter)" family = "claude-4" billing_policy = "anthropic" default = true -[models."anthropic/claude-sonnet-4-6".limits] +[providers.openrouter.models."claude-sonnet-4-6".limits] context_window = 1000000 max_output = 64000 -[models."anthropic/claude-sonnet-4-6".features] +[providers.openrouter.models."claude-sonnet-4-6".features] tools = true vision = true reasoning = true prompt_cache = true -[models."anthropic/claude-sonnet-4-6".costs] +[providers.openrouter.models."claude-sonnet-4-6".costs] input_cost_per_mtok = 3.0 output_cost_per_mtok = 15.0 cache_input_cost_per_mtok = 0.3 -[models."anthropic/claude-haiku-4-5"] -provider = "openrouter" +[providers.openrouter.models."claude-haiku-4-5"] api_id = "anthropic/claude-haiku-4.5" display_name = "Claude Haiku 4.5 (via OpenRouter)" family = "claude-4" billing_policy = "anthropic" small_default = true -[models."anthropic/claude-haiku-4-5".limits] +[providers.openrouter.models."claude-haiku-4-5".limits] context_window = 200000 max_output = 8192 -[models."anthropic/claude-haiku-4-5".features] +[providers.openrouter.models."claude-haiku-4-5".features] tools = true vision = true reasoning = false prompt_cache = true -[models."anthropic/claude-haiku-4-5".costs] +[providers.openrouter.models."claude-haiku-4-5".costs] input_cost_per_mtok = 1.0 output_cost_per_mtok = 5.0 cache_input_cost_per_mtok = 0.1 # ---------- OpenAI via OpenRouter ---------- -[models."openai/gpt-5.4"] -provider = "openrouter" +[providers.openrouter.models."gpt-5.4"] api_id = "openai/gpt-5.4" display_name = "GPT-5.4 (via OpenRouter)" family = "gpt-5" -[models."openai/gpt-5.4".limits] +[providers.openrouter.models."gpt-5.4".limits] context_window = 1050000 max_output = 32768 -[models."openai/gpt-5.4".features] +[providers.openrouter.models."gpt-5.4".features] tools = true vision = true reasoning = true -[models."openai/gpt-5.4".costs] +[providers.openrouter.models."gpt-5.4".costs] input_cost_per_mtok = 2.5 output_cost_per_mtok = 15.0 -[models."openai/gpt-5.5"] -provider = "openrouter" +[providers.openrouter.models."gpt-5.5"] api_id = "openai/gpt-5.5" display_name = "GPT-5.5 (via OpenRouter)" family = "gpt-5" -[models."openai/gpt-5.5".limits] +[providers.openrouter.models."gpt-5.5".limits] context_window = 1050000 max_output = 32768 -[models."openai/gpt-5.5".features] +[providers.openrouter.models."gpt-5.5".features] tools = true vision = true reasoning = true -[models."openai/gpt-5.5".costs] +[providers.openrouter.models."gpt-5.5".costs] input_cost_per_mtok = 5.0 output_cost_per_mtok = 30.0 # ---------- Google Gemini via OpenRouter ---------- -[models."google/gemini-3.1-pro-preview"] -provider = "openrouter" +[providers.openrouter.models."gemini-3.1-pro-preview"] api_id = "google/gemini-3.1-pro-preview" display_name = "Gemini 3.1 Pro Preview (via OpenRouter)" family = "gemini-3" -[models."google/gemini-3.1-pro-preview".limits] +[providers.openrouter.models."gemini-3.1-pro-preview".limits] context_window = 1048576 max_output = 65536 -[models."google/gemini-3.1-pro-preview".features] +[providers.openrouter.models."gemini-3.1-pro-preview".features] tools = true vision = true reasoning = true -[models."google/gemini-3.1-pro-preview".costs] +[providers.openrouter.models."gemini-3.1-pro-preview".costs] input_cost_per_mtok = 2.0 output_cost_per_mtok = 12.0 -[models."google/gemini-3.5-flash"] -provider = "openrouter" +[providers.openrouter.models."gemini-3.5-flash"] api_id = "google/gemini-3.5-flash" display_name = "Gemini 3.5 Flash (via OpenRouter)" family = "gemini-3" -[models."google/gemini-3.5-flash".limits] +[providers.openrouter.models."gemini-3.5-flash".limits] context_window = 1048576 max_output = 65536 -[models."google/gemini-3.5-flash".features] +[providers.openrouter.models."gemini-3.5-flash".features] tools = true vision = true reasoning = false -[models."google/gemini-3.5-flash".costs] +[providers.openrouter.models."gemini-3.5-flash".costs] input_cost_per_mtok = 1.5 output_cost_per_mtok = 9.0 # ---------- Open-weights models ---------- -[models."xiaomi/mimo-v2.5-pro"] -provider = "openrouter" +[providers.openrouter.models."mimo-v2.5-pro"] api_id = "xiaomi/mimo-v2.5-pro" display_name = "Xiaomi MiMo v2.5 Pro" family = "mimo-v2" -[models."xiaomi/mimo-v2.5-pro".limits] +[providers.openrouter.models."mimo-v2.5-pro".limits] context_window = 1050000 max_output = 16384 -[models."xiaomi/mimo-v2.5-pro".features] +[providers.openrouter.models."mimo-v2.5-pro".features] tools = true vision = false reasoning = false -[models."xiaomi/mimo-v2.5-pro".costs] +[providers.openrouter.models."mimo-v2.5-pro".costs] input_cost_per_mtok = 0.435 output_cost_per_mtok = 0.87 -[models."minimax/minimax-m2.7"] -provider = "openrouter" +[providers.openrouter.models."minimax-m2.7"] api_id = "minimax/minimax-m2.7" display_name = "MiniMax M2.7" family = "minimax-m2" -[models."minimax/minimax-m2.7".limits] +[providers.openrouter.models."minimax-m2.7".limits] context_window = 200000 max_output = 16384 -[models."minimax/minimax-m2.7".features] +[providers.openrouter.models."minimax-m2.7".features] tools = true vision = false reasoning = false -[models."minimax/minimax-m2.7".costs] +[providers.openrouter.models."minimax-m2.7".costs] input_cost_per_mtok = 0.28 output_cost_per_mtok = 1.20 -[models."deepseek/deepseek-v4-pro"] -provider = "openrouter" +[providers.openrouter.models."deepseek-v4-pro"] api_id = "deepseek/deepseek-v4-pro" display_name = "DeepSeek V4 Pro" family = "deepseek-v4" -[models."deepseek/deepseek-v4-pro".limits] +[providers.openrouter.models."deepseek-v4-pro".limits] context_window = 1050000 max_output = 16384 -[models."deepseek/deepseek-v4-pro".features] +[providers.openrouter.models."deepseek-v4-pro".features] tools = true vision = false reasoning = true -[models."deepseek/deepseek-v4-pro".costs] +[providers.openrouter.models."deepseek-v4-pro".costs] input_cost_per_mtok = 0.435 output_cost_per_mtok = 0.87 -[models."deepseek/deepseek-v4-flash"] -provider = "openrouter" +[providers.openrouter.models."deepseek-v4-flash"] api_id = "deepseek/deepseek-v4-flash" display_name = "DeepSeek V4 Flash" family = "deepseek-v4" -[models."deepseek/deepseek-v4-flash".limits] +[providers.openrouter.models."deepseek-v4-flash".limits] context_window = 1050000 max_output = 16384 -[models."deepseek/deepseek-v4-flash".features] +[providers.openrouter.models."deepseek-v4-flash".features] tools = true vision = false reasoning = false -[models."deepseek/deepseek-v4-flash".costs] +[providers.openrouter.models."deepseek-v4-flash".costs] input_cost_per_mtok = 0.10 output_cost_per_mtok = 0.20 -[models."moonshotai/kimi-k2.6"] -provider = "openrouter" +[providers.openrouter.models."kimi-k2.6"] api_id = "moonshotai/kimi-k2.6" display_name = "Kimi K2.6" family = "kimi-k2" -[models."moonshotai/kimi-k2.6".limits] +[providers.openrouter.models."kimi-k2.6".limits] context_window = 262144 max_output = 16384 -[models."moonshotai/kimi-k2.6".features] +[providers.openrouter.models."kimi-k2.6".features] tools = true vision = false reasoning = false -[models."moonshotai/kimi-k2.6".costs] +[providers.openrouter.models."kimi-k2.6".costs] input_cost_per_mtok = 0.73 output_cost_per_mtok = 3.49 -[models."moonshotai/kimi-k3"] -provider = "openrouter" +[providers.openrouter.models."kimi-k3"] api_id = "moonshotai/kimi-k3" display_name = "Kimi K3 (via OpenRouter)" family = "kimi-k3" -[models."moonshotai/kimi-k3".limits] +[providers.openrouter.models."kimi-k3".limits] context_window = 1048576 max_output = 131072 -[models."moonshotai/kimi-k3".features] +[providers.openrouter.models."kimi-k3".features] tools = true vision = true reasoning = true @@ -296,47 +283,45 @@ reasoning_effort = "always_adaptive" prompt_cache = true sampling_params = false -[models."moonshotai/kimi-k3".controls] +[providers.openrouter.models."kimi-k3".controls] reasoning_effort = ["low", "high", "max"] -[models."moonshotai/kimi-k3".costs] +[providers.openrouter.models."kimi-k3".costs] input_cost_per_mtok = 3.0 output_cost_per_mtok = 15.0 cache_input_cost_per_mtok = 0.3 -[models."poolside/laguna-s-2.1"] -provider = "openrouter" +[providers.openrouter.models."laguna-s-2.1"] api_id = "poolside/laguna-s-2.1" display_name = "Laguna S 2.1 (via OpenRouter)" family = "laguna-2" -[models."poolside/laguna-s-2.1".limits] +[providers.openrouter.models."laguna-s-2.1".limits] context_window = 1048576 max_output = 131072 -[models."poolside/laguna-s-2.1".features] +[providers.openrouter.models."laguna-s-2.1".features] tools = true vision = false reasoning = true prompt_cache = true sampling_params = true -[models."poolside/laguna-s-2.1".costs] +[providers.openrouter.models."laguna-s-2.1".costs] input_cost_per_mtok = 0.10 output_cost_per_mtok = 0.20 cache_input_cost_per_mtok = 0.01 -[models."poolside/laguna-xs-2.1"] -provider = "openrouter" +[providers.openrouter.models."laguna-xs-2.1"] api_id = "poolside/laguna-xs-2.1" display_name = "Laguna XS 2.1 (via OpenRouter)" family = "laguna-2" -[models."poolside/laguna-xs-2.1".limits] +[providers.openrouter.models."laguna-xs-2.1".limits] context_window = 262144 max_output = 32768 -[models."poolside/laguna-xs-2.1".features] +[providers.openrouter.models."laguna-xs-2.1".features] tools = true vision = false reasoning = true @@ -345,127 +330,121 @@ sampling_params = true # Current promotional rate. OpenRouter's authoritative in-band usage.cost # supersedes this estimate on completed responses. -[models."poolside/laguna-xs-2.1".costs] +[providers.openrouter.models."laguna-xs-2.1".costs] input_cost_per_mtok = 0.06 output_cost_per_mtok = 0.12 cache_input_cost_per_mtok = 0.03 -[models."qwen/qwen3-coder"] -provider = "openrouter" +[providers.openrouter.models."qwen3-coder"] api_id = "qwen/qwen3-coder" display_name = "Qwen3 Coder" family = "qwen3" -[models."qwen/qwen3-coder".limits] +[providers.openrouter.models."qwen3-coder".limits] context_window = 1050000 max_output = 16384 -[models."qwen/qwen3-coder".features] +[providers.openrouter.models."qwen3-coder".features] tools = true vision = false reasoning = false -[models."qwen/qwen3-coder".costs] +[providers.openrouter.models."qwen3-coder".costs] input_cost_per_mtok = 0.22 output_cost_per_mtok = 1.80 -[models."qwen/qwen3.6-flash"] -provider = "openrouter" +[providers.openrouter.models."qwen3.6-flash"] api_id = "qwen/qwen3.6-flash" display_name = "Qwen3.6 Flash" family = "qwen3" -[models."qwen/qwen3.6-flash".limits] +[providers.openrouter.models."qwen3.6-flash".limits] context_window = 1000000 max_output = 16384 -[models."qwen/qwen3.6-flash".features] +[providers.openrouter.models."qwen3.6-flash".features] tools = true vision = false reasoning = false -[models."qwen/qwen3.6-flash".costs] +[providers.openrouter.models."qwen3.6-flash".costs] input_cost_per_mtok = 0.1875 output_cost_per_mtok = 1.125 -[models."z-ai/glm-5.2"] -provider = "openrouter" +[providers.openrouter.models."glm-5.2"] api_id = "z-ai/glm-5.2" display_name = "GLM 5.2 (via OpenRouter)" family = "glm-5" -[models."z-ai/glm-5.2".limits] +[providers.openrouter.models."glm-5.2".limits] context_window = 1048576 max_output = 131072 -[models."z-ai/glm-5.2".features] +[providers.openrouter.models."glm-5.2".features] tools = true vision = false reasoning = true reasoning_effort = "levels" prompt_cache = true -[models."z-ai/glm-5.2".controls] +[providers.openrouter.models."glm-5.2".controls] reasoning_effort = ["high", "xhigh"] -[models."z-ai/glm-5.2".costs] +[providers.openrouter.models."glm-5.2".costs] input_cost_per_mtok = 0.784 output_cost_per_mtok = 2.464 cache_input_cost_per_mtok = 0.1456 -[models."z-ai/glm-4.6"] -provider = "openrouter" +[providers.openrouter.models."glm-4.6"] api_id = "z-ai/glm-4.6" display_name = "GLM 4.6" family = "glm-4" -[models."z-ai/glm-4.6".limits] +[providers.openrouter.models."glm-4.6".limits] context_window = 203000 max_output = 16384 -[models."z-ai/glm-4.6".features] +[providers.openrouter.models."glm-4.6".features] tools = true vision = false reasoning = false -[models."z-ai/glm-4.6".costs] +[providers.openrouter.models."glm-4.6".costs] input_cost_per_mtok = 0.43 output_cost_per_mtok = 1.74 -[models."nvidia/nemotron-3-super-120b-a12b"] -provider = "openrouter" +[providers.openrouter.models."nemotron-3-super"] api_id = "nvidia/nemotron-3-super-120b-a12b" display_name = "NVIDIA Nemotron 3 Super 120B" family = "nemotron-3" -[models."nvidia/nemotron-3-super-120b-a12b".limits] +[providers.openrouter.models."nemotron-3-super".limits] context_window = 1000000 max_output = 16384 -[models."nvidia/nemotron-3-super-120b-a12b".features] +[providers.openrouter.models."nemotron-3-super".features] tools = true vision = false reasoning = false -[models."nvidia/nemotron-3-super-120b-a12b".costs] +[providers.openrouter.models."nemotron-3-super".costs] input_cost_per_mtok = 0.09 output_cost_per_mtok = 0.45 -[models."mistralai/devstral-2512"] -provider = "openrouter" +[providers.openrouter.models."devstral-2"] api_id = "mistralai/devstral-2512" display_name = "Devstral 2512" family = "devstral" -[models."mistralai/devstral-2512".limits] +[providers.openrouter.models."devstral-2".limits] context_window = 262144 max_output = 16384 -[models."mistralai/devstral-2512".features] +[providers.openrouter.models."devstral-2".features] tools = true vision = false reasoning = false -[models."mistralai/devstral-2512".costs] +[providers.openrouter.models."devstral-2".costs] input_cost_per_mtok = 0.40 output_cost_per_mtok = 2.00 diff --git a/lib/crates/fabro-model/src/catalog/providers/poolside.toml b/lib/crates/fabro-model/src/catalog/providers/poolside.toml index 8fd1b6855..65ce87246 100644 --- a/lib/crates/fabro-model/src/catalog/providers/poolside.toml +++ b/lib/crates/fabro-model/src/catalog/providers/poolside.toml @@ -8,19 +8,18 @@ priority = 65 [providers.poolside.auth] credentials = ["env:POOLSIDE_API_KEY", "vault:POOLSIDE_API_KEY"] -[models."laguna-s-2.1"] -provider = "poolside" +[providers.poolside.models."laguna-s-2.1"] api_id = "poolside/laguna-s-2.1" display_name = "Laguna S 2.1" family = "laguna-2" default = true aliases = ["laguna", "laguna-s"] -[models."laguna-s-2.1".limits] +[providers.poolside.models."laguna-s-2.1".limits] context_window = 1048576 max_output = 131072 -[models."laguna-s-2.1".features] +[providers.poolside.models."laguna-s-2.1".features] tools = true vision = false reasoning = true @@ -30,13 +29,12 @@ sampling_params = true # Poolside Platform is free for a limited preview period. Keep the published # paid hosted rate as Fabro's durable estimate for paid access and post-preview # usage. -[models."laguna-s-2.1".costs] +[providers.poolside.models."laguna-s-2.1".costs] input_cost_per_mtok = 0.10 output_cost_per_mtok = 0.20 cache_input_cost_per_mtok = 0.01 -[models."laguna-xs-2.1"] -provider = "poolside" +[providers.poolside.models."laguna-xs-2.1"] api_id = "poolside/laguna-xs-2.1" display_name = "Laguna XS 2.1" family = "laguna-2" @@ -44,11 +42,11 @@ small_default = true probe = true aliases = ["laguna-xs"] -[models."laguna-xs-2.1".limits] +[providers.poolside.models."laguna-xs-2.1".limits] context_window = 262144 max_output = 32768 -[models."laguna-xs-2.1".features] +[providers.poolside.models."laguna-xs-2.1".features] tools = true vision = false reasoning = true @@ -57,7 +55,7 @@ sampling_params = true # Poolside Platform is free for a limited preview period. These are Poolside's # published paid endpoint rates. -[models."laguna-xs-2.1".costs] +[providers.poolside.models."laguna-xs-2.1".costs] input_cost_per_mtok = 0.10 output_cost_per_mtok = 0.20 cache_input_cost_per_mtok = 0.05 diff --git a/lib/crates/fabro-model/src/catalog/providers/venice.toml b/lib/crates/fabro-model/src/catalog/providers/venice.toml index bb91aeada..dc97c4ab4 100644 --- a/lib/crates/fabro-model/src/catalog/providers/venice.toml +++ b/lib/crates/fabro-model/src/catalog/providers/venice.toml @@ -8,43 +8,39 @@ aliases = ["venice-ai"] [providers.venice.auth] credentials = ["env:VENICE_API_KEY", "vault:VENICE_API_KEY"] -[models."venice-uncensored-1-2"] -provider = "venice" -api_id = "venice-uncensored-1-2" +[providers.venice.models."venice-uncensored-1-2"] display_name = "Venice Uncensored 1.2" family = "venice-uncensored" default = true aliases = ["venice-uncensored", "vu"] -[models."venice-uncensored-1-2".limits] +[providers.venice.models."venice-uncensored-1-2".limits] context_window = 128000 max_output = 8192 -[models."venice-uncensored-1-2".features] +[providers.venice.models."venice-uncensored-1-2".features] tools = true vision = true reasoning = false -[models."venice-uncensored-1-2".costs] +[providers.venice.models."venice-uncensored-1-2".costs] input_cost_per_mtok = 0.2 output_cost_per_mtok = 0.9 -[models."venice-uncensored-role-play"] -provider = "venice" -api_id = "venice-uncensored-role-play" +[providers.venice.models."venice-uncensored-role-play"] display_name = "Venice Uncensored Role Play" family = "venice-uncensored" aliases = ["venice-roleplay", "vrp"] -[models."venice-uncensored-role-play".limits] +[providers.venice.models."venice-uncensored-role-play".limits] context_window = 128000 max_output = 4096 -[models."venice-uncensored-role-play".features] +[providers.venice.models."venice-uncensored-role-play".features] tools = true vision = true reasoning = false -[models."venice-uncensored-role-play".costs] +[providers.venice.models."venice-uncensored-role-play".costs] input_cost_per_mtok = 0.5 output_cost_per_mtok = 2.0 diff --git a/lib/crates/fabro-model/src/catalog/providers/zai.toml b/lib/crates/fabro-model/src/catalog/providers/zai.toml index e720df4da..c6d70cd64 100644 --- a/lib/crates/fabro-model/src/catalog/providers/zai.toml +++ b/lib/crates/fabro-model/src/catalog/providers/zai.toml @@ -8,50 +8,46 @@ priority = 60 [providers.zai.auth] credentials = ["env:ZAI_API_KEY", "vault:ZAI_API_KEY"] -[models."glm-5.2"] -provider = "zai" -api_id = "glm-5.2" +[providers.zai.models."glm-5.2"] display_name = "GLM 5.2" family = "glm-5" default = true aliases = ["glm", "glm5"] -[models."glm-5.2".limits] +[providers.zai.models."glm-5.2".limits] context_window = 1048576 max_output = 131072 -[models."glm-5.2".features] +[providers.zai.models."glm-5.2".features] tools = true vision = false reasoning = true reasoning_effort = "levels" prompt_cache = true -[models."glm-5.2".controls] +[providers.zai.models."glm-5.2".controls] reasoning_effort = ["high", "max"] -[models."glm-5.2".costs] +[providers.zai.models."glm-5.2".costs] input_cost_per_mtok = 1.4 output_cost_per_mtok = 4.4 cache_input_cost_per_mtok = 0.26 -[models."glm-4.7"] -provider = "zai" -api_id = "glm-4.7" +[providers.zai.models."glm-4.7"] display_name = "GLM 4.7" family = "glm-4" estimated_output_tps = 100 aliases = ["glm4"] -[models."glm-4.7".limits] +[providers.zai.models."glm-4.7".limits] context_window = 202752 max_output = 16384 -[models."glm-4.7".features] +[providers.zai.models."glm-4.7".features] tools = true vision = false reasoning = false -[models."glm-4.7".costs] +[providers.zai.models."glm-4.7".costs] input_cost_per_mtok = 0.6 output_cost_per_mtok = 2.2 diff --git a/lib/crates/fabro-model/src/ids.rs b/lib/crates/fabro-model/src/ids.rs index 18d1bdd86..94b6f043e 100644 --- a/lib/crates/fabro-model/src/ids.rs +++ b/lib/crates/fabro-model/src/ids.rs @@ -101,9 +101,12 @@ impl AsRef for ProviderId { } } -/// Stable model identifier — either the canonical catalog ID or one of its -/// declared aliases. -#[derive(Debug, Clone, PartialEq, Eq, Hash, PartialOrd, Ord, Serialize, Deserialize)] +/// Stable, canonical human-facing model slug. +/// +/// Aliases are selectors that resolve to a `ModelId`; they are never model +/// IDs themselves. The same canonical slug may identify one offering on each +/// provider, so an offering's full identity is `(ProviderId, ModelId)`. +#[derive(Clone, PartialEq, Eq, Hash, PartialOrd, Ord, Serialize, Deserialize)] #[serde(transparent)] pub struct ModelId(String); @@ -123,12 +126,32 @@ impl ModelId { } } +impl std::borrow::Borrow for ModelId { + fn borrow(&self) -> &str { + self.as_str() + } +} + +impl std::ops::Deref for ModelId { + type Target = str; + + fn deref(&self) -> &Self::Target { + self.as_str() + } +} + impl fmt::Display for ModelId { fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { f.write_str(&self.0) } } +impl fmt::Debug for ModelId { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + fmt::Debug::fmt(&self.0, f) + } +} + impl From<&str> for ModelId { fn from(s: &str) -> Self { Self(s.to_string()) @@ -147,6 +170,30 @@ impl AsRef for ModelId { } } +impl PartialEq for ModelId { + fn eq(&self, other: &str) -> bool { + self.as_str() == other + } +} + +impl PartialEq<&str> for ModelId { + fn eq(&self, other: &&str) -> bool { + self.as_str() == *other + } +} + +impl PartialEq for str { + fn eq(&self, other: &ModelId) -> bool { + self == other.as_str() + } +} + +impl PartialEq for &str { + fn eq(&self, other: &ModelId) -> bool { + *self == other.as_str() + } +} + #[cfg(test)] mod tests { use super::*; diff --git a/lib/crates/fabro-model/src/types.rs b/lib/crates/fabro-model/src/types.rs index 8dcbd3270..312067f9a 100644 --- a/lib/crates/fabro-model/src/types.rs +++ b/lib/crates/fabro-model/src/types.rs @@ -1,6 +1,6 @@ use serde::{Deserialize, Serialize}; -use crate::ids::ProviderId; +use crate::ids::{ModelId, ProviderId}; // --- 2.9 Model --- @@ -78,7 +78,7 @@ pub struct ModelCosts { #[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] pub struct Model { - pub id: String, + pub id: ModelId, pub provider: ProviderId, pub family: String, pub display_name: String, @@ -210,7 +210,7 @@ mod tests { #[test] fn inherent_methods_return_correct_values() { let info = Model { - id: "model-id".to_string(), + id: ModelId::new("model-id"), provider: ProviderId::new("provider-id"), family: "family".to_string(), display_name: "Display Name".to_string(),