diff --git a/Cargo.lock b/Cargo.lock
index b8d25cf7a..dd03ef46e 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -12,11 +12,11 @@ dependencies = [
  "glob",
  "jsonschema",
  "libc",
+ "llm",
  "serde",
  "serde_json",
  "thiserror 2.0.18",
  "tokio",
- "unified-llm",
  "uuid",
 ]
 
@@ -159,6 +159,7 @@ dependencies = [
  "dotenvy",
  "futures",
  "http-body-util",
+ "llm",
  "nom",
  "rand",
  "serde",
@@ -168,7 +169,6 @@ dependencies = [
  "tokio",
  "tokio-stream",
  "tower",
- "unified-llm",
  "uuid",
 ]
 
@@ -1188,6 +1188,31 @@ version = "0.8.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6373607a59f0be73a39b6fe456b8192fcc3585f602af20751600e974dd455e77"
 
+[[package]]
+name = "llm"
+version = "0.1.0"
+dependencies = [
+ "anyhow",
+ "assert_cmd",
+ "async-trait",
+ "base64",
+ "bytes",
+ "clap",
+ "dotenvy",
+ "futures",
+ "http",
+ "predicates",
+ "rand",
+ "reqwest 0.12.28",
+ "serde",
+ "serde_json",
+ "thiserror 2.0.18",
+ "tokio",
+ "tokio-stream",
+ "tokio-util",
+ "uuid",
+]
+
 [[package]]
 name = "lock_api"
 version = "0.4.14"
@@ -2298,31 +2323,6 @@ version = "0.2.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ebc1c04c71510c7f702b52b7c350734c9ff1295c464a03335b00bb84fc54f853"
 
-[[package]]
-name = "unified-llm"
-version = "0.1.0"
-dependencies = [
- "anyhow",
- "assert_cmd",
- "async-trait",
- "base64",
- "bytes",
- "clap",
- "dotenvy",
- "futures",
- "http",
- "predicates",
- "rand",
- "reqwest 0.12.28",
- "serde",
- "serde_json",
- "thiserror 2.0.18",
- "tokio",
- "tokio-stream",
- "tokio-util",
- "uuid",
-]
-
 [[package]]
 name = "untrusted"
 version = "0.9.0"
diff --git a/crates/agent/Cargo.toml b/crates/agent/Cargo.toml
index 1bbfd422d..a2e4858cd 100644
--- a/crates/agent/Cargo.toml
+++ b/crates/agent/Cargo.toml
@@ -10,7 +10,7 @@ keywords = ["llm", "ai", "agent", "coding"]
 categories = ["api-bindings"]
 
 [dependencies]
-unified-llm = { path = "../unified-llm" }
+llm = { path = "../llm" }
 thiserror.workspace = true
 serde.workspace = true
 serde_json.workspace = true
diff --git a/crates/agent/src/error.rs b/crates/agent/src/error.rs
index 7af71778f..a3fb690bc 100644
--- a/crates/agent/src/error.rs
+++ b/crates/agent/src/error.rs
@@ -1,4 +1,4 @@
-use unified_llm::error::SdkError;
+use llm::error::SdkError;
 
 #[derive(Debug, thiserror::Error)]
 pub enum AgentError {
diff --git a/crates/agent/src/history.rs b/crates/agent/src/history.rs
index e15a0f37b..215b6c6d7 100644
--- a/crates/agent/src/history.rs
+++ b/crates/agent/src/history.rs
@@ -1,5 +1,5 @@
 use crate::types::Turn;
-use unified_llm::types::{ContentPart, Message, Role};
+use llm::types::{ContentPart, Message, Role};
 
 #[derive(Debug, Clone, Default)]
 pub struct History {
@@ -29,7 +29,7 @@ impl History {
                     let mut parts: Vec<ContentPart> = Vec::new();
                     if let Some(reasoning_text) = reasoning {
                         parts.push(ContentPart::Thinking(
-                            unified_llm::types::ThinkingData {
+                            llm::types::ThinkingData {
                                 text: reasoning_text.clone(),
                                 signature: None,
                                 redacted: false,
@@ -79,7 +79,7 @@ impl History {
 mod tests {
     use super::*;
     use std::time::SystemTime;
-    use unified_llm::types::{ToolCall, ToolResult, Usage};
+    use llm::types::{ToolCall, ToolResult, Usage};
 
     #[test]
     fn empty_history_produces_empty_messages() {
diff --git a/crates/agent/src/loop_detection.rs b/crates/agent/src/loop_detection.rs
index 70c19c632..c48e14bc9 100644
--- a/crates/agent/src/loop_detection.rs
+++ b/crates/agent/src/loop_detection.rs
@@ -94,7 +94,7 @@ fn is_repeating_pattern(signatures: &[u64], pattern_len: usize) -> bool {
 mod tests {
     use super::*;
     use std::time::SystemTime;
-    use unified_llm::types::{ToolCall, Usage};
+    use llm::types::{ToolCall, Usage};
 
     fn assistant_with_tool(name: &str, args: serde_json::Value) -> Turn {
         Turn::Assistant {
diff --git a/crates/agent/src/profiles/openai.rs b/crates/agent/src/profiles/openai.rs
index 95a31a153..c5c53efc2 100644
--- a/crates/agent/src/profiles/openai.rs
+++ b/crates/agent/src/profiles/openai.rs
@@ -3,7 +3,7 @@ use crate::profiles::assemble_system_prompt;
 use crate::profiles::BaseProfile;
 use crate::provider_profile::{ProfileCapabilities, ProviderProfile};
 use crate::tool_registry::{RegisteredTool, ToolRegistry};
-use unified_llm::types::ToolDefinition;
+use llm::types::ToolDefinition;
 use crate::tools::{
     make_glob_tool, make_grep_tool, make_read_file_tool, make_shell_tool, make_write_file_tool,
 };
diff --git a/crates/agent/src/provider_profile.rs b/crates/agent/src/provider_profile.rs
index 8bc27ff60..9c2e8b42a 100644
--- a/crates/agent/src/provider_profile.rs
+++ b/crates/agent/src/provider_profile.rs
@@ -6,7 +6,7 @@ use crate::subagent::{
 };
 use crate::tool_registry::ToolRegistry;
 use std::sync::Arc;
-use unified_llm::types::ToolDefinition;
+use llm::types::ToolDefinition;
 
 /// Static capabilities of a provider profile.
 pub struct ProfileCapabilities {
diff --git a/crates/agent/src/session.rs b/crates/agent/src/session.rs
index 7c755e421..9083bf1de 100644
--- a/crates/agent/src/session.rs
+++ b/crates/agent/src/session.rs
@@ -14,9 +14,9 @@ use std::collections::VecDeque;
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::{Arc, Mutex};
 use std::time::SystemTime;
-use unified_llm::client::Client;
-use unified_llm::error::{ProviderErrorKind, SdkError};
-use unified_llm::types::{Message, Request, ToolChoice, ToolResult};
+use llm::client::Client;
+use llm::error::{ProviderErrorKind, SdkError};
+use llm::types::{Message, Request, ToolChoice, ToolResult};
 
 pub struct Session {
     id: String,
@@ -398,7 +398,7 @@ impl Session {
 
     async fn execute_tool_calls(
         &mut self,
-        tool_calls: &[unified_llm::types::ToolCall],
+        tool_calls: &[llm::types::ToolCall],
     ) -> Vec<ToolResult> {
         if self.provider_profile.supports_parallel_tool_calls() && tool_calls.len() > 1 {
             self.execute_tool_calls_parallel(tool_calls).await
@@ -409,7 +409,7 @@ impl Session {
 
     async fn execute_tool_calls_sequential(
         &self,
-        tool_calls: &[unified_llm::types::ToolCall],
+        tool_calls: &[llm::types::ToolCall],
     ) -> Vec<ToolResult> {
         let mut results = Vec::new();
         for tc in tool_calls {
@@ -450,7 +450,7 @@ impl Session {
 
     async fn execute_tool_calls_parallel(
         &self,
-        tool_calls: &[unified_llm::types::ToolCall],
+        tool_calls: &[llm::types::ToolCall],
     ) -> Vec<ToolResult> {
         let emitter = self.event_emitter.clone();
         let env = self.execution_env.clone();
@@ -669,9 +669,9 @@ mod tests {
     use super::*;
     use crate::test_support::*;
     use crate::tool_registry::{RegisteredTool, ToolRegistry};
-    use unified_llm::error::ProviderErrorDetail;
-    use unified_llm::provider::ProviderAdapter;
-    use unified_llm::types::ToolDefinition;
+    use llm::error::ProviderErrorDetail;
+    use llm::provider::ProviderAdapter;
+    use llm::types::ToolDefinition;
 
     // --- Tests ---
 
diff --git a/crates/agent/src/subagent.rs b/crates/agent/src/subagent.rs
index 5015f5ca0..934e1d06c 100644
--- a/crates/agent/src/subagent.rs
+++ b/crates/agent/src/subagent.rs
@@ -6,7 +6,7 @@ use crate::types::Turn;
 use std::collections::{HashMap, VecDeque};
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::{Arc, Mutex};
-use unified_llm::types::ToolDefinition;
+use llm::types::ToolDefinition;
 
 pub type SessionFactory = Arc<dyn Fn() -> Session + Send + Sync>;
 
diff --git a/crates/agent/src/test_support.rs b/crates/agent/src/test_support.rs
index b9100cd45..8d7d58d98 100644
--- a/crates/agent/src/test_support.rs
+++ b/crates/agent/src/test_support.rs
@@ -8,10 +8,10 @@ use async_trait::async_trait;
 use std::collections::HashMap;
 use std::sync::atomic::{AtomicUsize, Ordering};
 use std::sync::{Arc, Mutex};
-use unified_llm::client::Client;
-use unified_llm::error::SdkError;
-use unified_llm::provider::{ProviderAdapter, StreamEventStream};
-use unified_llm::types::{FinishReason, Message, Request, Response, Usage};
+use llm::client::Client;
+use llm::error::SdkError;
+use llm::provider::{ProviderAdapter, StreamEventStream};
+use llm::types::{FinishReason, Message, Request, Response, Usage};
 
 // --- MockExecutionEnvironment ---
 
@@ -474,7 +474,7 @@ pub(crate) fn tool_call_response(
     tool_call_id: &str,
     args: serde_json::Value,
 ) -> Response {
-    use unified_llm::types::{ContentPart, Role, ToolCall};
+    use llm::types::{ContentPart, Role, ToolCall};
     Response {
         id: format!("resp_{tool_call_id}"),
         model: "mock-model".into(),
@@ -502,7 +502,7 @@ pub(crate) fn tool_call_response(
 }
 
 pub(crate) fn make_echo_tool() -> crate::tool_registry::RegisteredTool {
-    use unified_llm::types::ToolDefinition;
+    use llm::types::ToolDefinition;
     crate::tool_registry::RegisteredTool {
         definition: ToolDefinition {
             name: "echo".into(),
@@ -522,7 +522,7 @@ pub(crate) fn make_echo_tool() -> crate::tool_registry::RegisteredTool {
 }
 
 pub(crate) fn make_error_tool() -> crate::tool_registry::RegisteredTool {
-    use unified_llm::types::ToolDefinition;
+    use llm::types::ToolDefinition;
     crate::tool_registry::RegisteredTool {
         definition: ToolDefinition {
             name: "fail_tool".into(),
@@ -597,7 +597,7 @@ impl ProviderAdapter for CapturingLlmProvider {
 pub(crate) fn multi_tool_call_response(
     calls: Vec<(&str, &str, serde_json::Value)>,
 ) -> Response {
-    use unified_llm::types::{ContentPart, Role, ToolCall};
+    use llm::types::{ContentPart, Role, ToolCall};
     let mut content = vec![ContentPart::text("Let me use multiple tools.")];
     for (tool_name, tool_call_id, args) in &calls {
         content.push(ContentPart::ToolCall(ToolCall::new(
diff --git a/crates/agent/src/tool_registry.rs b/crates/agent/src/tool_registry.rs
index c62d28da6..b944e83dd 100644
--- a/crates/agent/src/tool_registry.rs
+++ b/crates/agent/src/tool_registry.rs
@@ -3,7 +3,7 @@ use std::collections::HashMap;
 use std::future::Future;
 use std::pin::Pin;
 use std::sync::Arc;
-use unified_llm::types::ToolDefinition;
+use llm::types::ToolDefinition;
 
 pub type ToolExecutor = Arc<
     dyn Fn(
diff --git a/crates/agent/src/tools.rs b/crates/agent/src/tools.rs
index a23f0d3e1..276fecbd1 100644
--- a/crates/agent/src/tools.rs
+++ b/crates/agent/src/tools.rs
@@ -3,7 +3,7 @@ use crate::execution_env::GrepOptions;
 use crate::tool_registry::RegisteredTool;
 use std::fmt::Write;
 use std::sync::Arc;
-use unified_llm::types::ToolDefinition;
+use llm::types::ToolDefinition;
 
 pub(crate) fn required_str<'a>(args: &'a serde_json::Value, key: &str) -> Result<&'a str, String> {
     args.get(key)
diff --git a/crates/agent/src/types.rs b/crates/agent/src/types.rs
index a4cb0bd7c..7b329edbf 100644
--- a/crates/agent/src/types.rs
+++ b/crates/agent/src/types.rs
@@ -1,5 +1,5 @@
 use std::time::SystemTime;
-use unified_llm::types::{ToolCall, ToolResult, Usage};
+use llm::types::{ToolCall, ToolResult, Usage};
 
 #[derive(Debug, Clone)]
 pub enum Turn {
diff --git a/crates/attractor/Cargo.toml b/crates/attractor/Cargo.toml
index 66ded992f..63afd5611 100644
--- a/crates/attractor/Cargo.toml
+++ b/crates/attractor/Cargo.toml
@@ -22,7 +22,7 @@ clap.workspace = true
 anyhow.workspace = true
 dotenvy.workspace = true
 agent = { path = "../agent" }
-unified-llm = { path = "../unified-llm" }
+llm = { path = "../llm" }
 thiserror.workspace = true
 serde.workspace = true
 serde_json.workspace = true
diff --git a/crates/attractor/src/cli/backend.rs b/crates/attractor/src/cli/backend.rs
index bfddd581a..8ff96dc1a 100644
--- a/crates/attractor/src/cli/backend.rs
+++ b/crates/attractor/src/cli/backend.rs
@@ -7,7 +7,7 @@ use agent::{
     AnthropicProfile, GeminiProfile, LocalExecutionEnvironment, OpenAiProfile, ProviderProfile,
     Session, SessionConfig, Turn,
 };
-use unified_llm::client::Client;
+use llm::client::Client;
 
 use crate::context::Context;
 use crate::error::AttractorError;
diff --git a/crates/attractor/src/cli/run.rs b/crates/attractor/src/cli/run.rs
index bb3154a1d..fcbd51a4b 100644
--- a/crates/attractor/src/cli/run.rs
+++ b/crates/attractor/src/cli/run.rs
@@ -79,7 +79,7 @@ pub async fn run_command(args: RunArgs) -> anyhow::Result<()> {
     let dry_run_mode = if args.dry_run {
         true
     } else {
-        match unified_llm::client::Client::from_env().await {
+        match llm::client::Client::from_env().await {
             Ok(c) if c.provider_names().is_empty() => {
                 eprintln!("Warning: No LLM providers configured. Running in dry-run mode.");
                 true
diff --git a/crates/attractor/tests/integration.rs b/crates/attractor/tests/integration.rs
index 1452b819e..ae5257cf4 100644
--- a/crates/attractor/tests/integration.rs
+++ b/crates/attractor/tests/integration.rs
@@ -3600,8 +3600,8 @@ mod real_llm {
     use attractor::graph::Node;
     use attractor::handler::codergen::{CodergenBackend, CodergenHandler, CodergenResult};
 
-    use unified_llm::client::Client;
-    use unified_llm::types::{Message, Request};
+    use llm::client::Client;
+    use llm::types::{Message, Request};
 
     struct LlmCodergenBackend {
         client: Arc<Client>,
diff --git a/crates/unified-llm/Cargo.toml b/crates/llm/Cargo.toml
similarity index 97%
rename from crates/unified-llm/Cargo.toml
rename to crates/llm/Cargo.toml
index ed5bc6096..cbf38072e 100644
--- a/crates/unified-llm/Cargo.toml
+++ b/crates/llm/Cargo.toml
@@ -1,5 +1,5 @@
 [package]
-name = "unified-llm"
+name = "llm"
 edition.workspace = true
 version.workspace = true
 license.workspace = true
diff --git a/crates/unified-llm/README.md b/crates/llm/README.md
similarity index 100%
rename from crates/unified-llm/README.md
rename to crates/llm/README.md
diff --git a/crates/unified-llm/src/bin/ullm.rs b/crates/llm/src/bin/ullm.rs
similarity index 97%
rename from crates/unified-llm/src/bin/ullm.rs
rename to crates/llm/src/bin/ullm.rs
index 5e856e991..b3debcb1c 100644
--- a/crates/unified-llm/src/bin/ullm.rs
+++ b/crates/llm/src/bin/ullm.rs
@@ -3,8 +3,8 @@ use std::io::{self, IsTerminal, Read};
 use anyhow::{bail, Context, Result};
 use clap::{Parser, Subcommand};
 use futures::StreamExt;
-use unified_llm::catalog;
-use unified_llm::generate::{self, GenerateParams};
+use llm::catalog;
+use llm::generate::{self, GenerateParams};
 
 #[derive(Parser)]
 #[command(name = "ullm")]
@@ -69,7 +69,7 @@ fn parse_option(s: &str) -> Result<(String, String), String> {
     Ok((key.to_string(), value.to_string()))
 }
 
-fn print_models_table(models: &[unified_llm::types::ModelInfo]) {
+fn print_models_table(models: &[llm::types::ModelInfo]) {
     println!(
         "{:<30} {:<12} {:<30} {:>14}",
         "ID", "PROVIDER", "ALIASES", "CONTEXT"
@@ -168,7 +168,7 @@ struct PromptArgs {
     option: Vec<(String, String)>,
 }
 
-fn print_usage(usage: &unified_llm::types::Usage) {
+fn print_usage(usage: &llm::types::Usage) {
     eprintln!(
         "Tokens: {} input, {} output, {} total",
         usage.input_tokens, usage.output_tokens, usage.total_tokens
@@ -198,7 +198,7 @@ async fn run_prompt(args: PromptArgs) -> Result<()> {
     } else {
         let mut stream_result = generate::stream(params).await?;
         while let Some(event) = stream_result.next().await {
-            if let unified_llm::types::StreamEvent::TextDelta { delta, .. } = event? {
+            if let llm::types::StreamEvent::TextDelta { delta, .. } = event? {
                 print!("{delta}");
             }
         }
diff --git a/crates/unified-llm/src/catalog.json b/crates/llm/src/catalog.json
similarity index 100%
rename from crates/unified-llm/src/catalog.json
rename to crates/llm/src/catalog.json
diff --git a/crates/unified-llm/src/catalog.rs b/crates/llm/src/catalog.rs
similarity index 100%
rename from crates/unified-llm/src/catalog.rs
rename to crates/llm/src/catalog.rs
diff --git a/crates/unified-llm/src/client.rs b/crates/llm/src/client.rs
similarity index 100%
rename from crates/unified-llm/src/client.rs
rename to crates/llm/src/client.rs
diff --git a/crates/unified-llm/src/error.rs b/crates/llm/src/error.rs
similarity index 100%
rename from crates/unified-llm/src/error.rs
rename to crates/llm/src/error.rs
diff --git a/crates/unified-llm/src/generate.rs b/crates/llm/src/generate.rs
similarity index 100%
rename from crates/unified-llm/src/generate.rs
rename to crates/llm/src/generate.rs
diff --git a/crates/unified-llm/src/lib.rs b/crates/llm/src/lib.rs
similarity index 100%
rename from crates/unified-llm/src/lib.rs
rename to crates/llm/src/lib.rs
diff --git a/crates/unified-llm/src/middleware.rs b/crates/llm/src/middleware.rs
similarity index 100%
rename from crates/unified-llm/src/middleware.rs
rename to crates/llm/src/middleware.rs
diff --git a/crates/unified-llm/src/provider.rs b/crates/llm/src/provider.rs
similarity index 100%
rename from crates/unified-llm/src/provider.rs
rename to crates/llm/src/provider.rs
diff --git a/crates/unified-llm/src/providers/anthropic.rs b/crates/llm/src/providers/anthropic.rs
similarity index 100%
rename from crates/unified-llm/src/providers/anthropic.rs
rename to crates/llm/src/providers/anthropic.rs
diff --git a/crates/unified-llm/src/providers/common.rs b/crates/llm/src/providers/common.rs
similarity index 100%
rename from crates/unified-llm/src/providers/common.rs
rename to crates/llm/src/providers/common.rs
diff --git a/crates/unified-llm/src/providers/gemini.rs b/crates/llm/src/providers/gemini.rs
similarity index 100%
rename from crates/unified-llm/src/providers/gemini.rs
rename to crates/llm/src/providers/gemini.rs
diff --git a/crates/unified-llm/src/providers/mod.rs b/crates/llm/src/providers/mod.rs
similarity index 100%
rename from crates/unified-llm/src/providers/mod.rs
rename to crates/llm/src/providers/mod.rs
diff --git a/crates/unified-llm/src/providers/openai.rs b/crates/llm/src/providers/openai.rs
similarity index 100%
rename from crates/unified-llm/src/providers/openai.rs
rename to crates/llm/src/providers/openai.rs
diff --git a/crates/unified-llm/src/providers/openai_compatible.rs b/crates/llm/src/providers/openai_compatible.rs
similarity index 100%
rename from crates/unified-llm/src/providers/openai_compatible.rs
rename to crates/llm/src/providers/openai_compatible.rs
diff --git a/crates/unified-llm/src/retry.rs b/crates/llm/src/retry.rs
similarity index 100%
rename from crates/unified-llm/src/retry.rs
rename to crates/llm/src/retry.rs
diff --git a/crates/unified-llm/src/tools.rs b/crates/llm/src/tools.rs
similarity index 100%
rename from crates/unified-llm/src/tools.rs
rename to crates/llm/src/tools.rs
diff --git a/crates/unified-llm/src/types.rs b/crates/llm/src/types.rs
similarity index 100%
rename from crates/unified-llm/src/types.rs
rename to crates/llm/src/types.rs
diff --git a/crates/unified-llm/tests/integration.rs b/crates/llm/tests/integration.rs
similarity index 84%
rename from crates/unified-llm/tests/integration.rs
rename to crates/llm/tests/integration.rs
index 01c58f746..6ed2bae78 100644
--- a/crates/unified-llm/tests/integration.rs
+++ b/crates/llm/tests/integration.rs
@@ -1,6 +1,6 @@
-use unified_llm::provider::ProviderAdapter;
-use unified_llm::providers::{AnthropicAdapter, GeminiAdapter, OpenAiAdapter};
-use unified_llm::types::{Message, Request};
+use llm::provider::ProviderAdapter;
+use llm::providers::{AnthropicAdapter, GeminiAdapter, OpenAiAdapter};
+use llm::types::{Message, Request};
 
 fn make_request(model: &str) -> Request {
     Request {
@@ -30,7 +30,7 @@ async fn anthropic_complete() {
     let response = adapter.complete(&request).await.unwrap();
 
     assert!(!response.text().is_empty(), "response text should not be empty");
-    assert_eq!(response.finish_reason, unified_llm::types::FinishReason::Stop);
+    assert_eq!(response.finish_reason, llm::types::FinishReason::Stop);
     assert!(response.usage.input_tokens > 0);
     assert!(response.usage.output_tokens > 0);
     assert_eq!(response.provider, "anthropic");
@@ -46,7 +46,7 @@ async fn openai_complete() {
     let response = adapter.complete(&request).await.unwrap();
 
     assert!(!response.text().is_empty(), "response text should not be empty");
-    assert_eq!(response.finish_reason, unified_llm::types::FinishReason::Stop);
+    assert_eq!(response.finish_reason, llm::types::FinishReason::Stop);
     assert!(response.usage.input_tokens > 0);
     assert!(response.usage.output_tokens > 0);
     assert_eq!(response.provider, "openai");
@@ -62,7 +62,7 @@ async fn gemini_complete() {
     let response = adapter.complete(&request).await.unwrap();
 
     assert!(!response.text().is_empty(), "response text should not be empty");
-    assert_eq!(response.finish_reason, unified_llm::types::FinishReason::Stop);
+    assert_eq!(response.finish_reason, llm::types::FinishReason::Stop);
     assert!(response.usage.input_tokens > 0);
     assert!(response.usage.output_tokens > 0);
     assert_eq!(response.provider, "gemini");