Enable OpenAI Responses background mode with resilient streaming for GPT‑5 Pro and any model flagged via metadata.
Key changes:
- Background mode enablement
• Auto-enable for models with info.backgroundMode === true (e.g., gpt-5-pro-2025-10-06) defined in [packages/types/src/providers/openai.ts](packages/types/src/providers/openai.ts).
• Also respects manual override (openAiNativeBackgroundMode) from ProviderSettings/ApiHandlerOptions.
- Request shape (Responses API)
• background:true, stream:true, store:true set in [OpenAiNativeHandler.buildRequestBody()](src/api/providers/openai-native.ts:224).
- Streaming UX and status events
• New ApiStreamStatusChunk in [src/api/transform/stream.ts](src/api/transform/stream.ts) with statuses: queued, in_progress, completed, failed, canceled, reconnecting, polling.
• Provider emits status chunks in SDK + SSE paths via [OpenAiNativeHandler.processEvent()](src/api/providers/openai-native.ts:1100) and [OpenAiNativeHandler.handleStreamResponse()](src/api/providers/openai-native.ts:651).
• UI spinner shows background lifecycle labels in [webview-ui/src/components/chat/ChatRow.tsx](webview-ui/src/components/chat/ChatRow.tsx) using [webview-ui/src/utils/backgroundStatus.ts](webview-ui/src/utils/backgroundStatus.ts).
- Resilience: auto-resume + poll fallback
• On stream drop for background tasks, attempt SSE resume using response.id and last sequence_number with exponential backoff in [OpenAiNativeHandler.attemptResumeOrPoll()](src/api/providers/openai-native.ts:1215).
• If resume fails, poll GET /v1/responses/{id} every 2s until terminal and synthesize final output/usage.
• Deduplicate resumed events via resumeCutoffSequence in [handleStreamResponse()](src/api/providers/openai-native.ts:737).
- Settings (no new UI switch)
• Added optional provider settings and ApiHandlerOptions: autoResume, resumeMaxRetries, resumeBaseDelayMs, pollIntervalMs, pollMaxMinutes in [packages/types/src/provider-settings.ts](packages/types/src/provider-settings.ts) and [src/shared/api.ts](src/shared/api.ts).
- Cleanup
• Removed VS Code contributes toggle for background mode; behavior now model-driven + programmatic override.
- Tests
• Provider: coverage for background status emission, auto-resume success, resume→poll fallback, non-background negative in [src/api/providers/__tests__/openai-native.spec.ts](src/api/providers/__tests__/openai-native.spec.ts).
• Usage parity unchanged validated in [src/api/providers/__tests__/openai-native-usage.spec.ts](src/api/providers/__tests__/openai-native-usage.spec.ts).
• UI: label mapping tests for background statuses in [webview-ui/src/utils/__tests__/backgroundStatus.spec.ts](webview-ui/src/utils/__tests__/backgroundStatus.spec.ts).
Notes:
- Aligns with TEMP_OPENAI_BACKGROUND_TASK_DOCS.DM: background requires store=true; supports streaming resume via response.id + sequence_number.
- Default behavior unchanged for non-background models; no breaking changes.
- Add GPT‑5 Pro to model registry with:
- contextWindow: 400k, maxTokens: 272k
- supportsImages: true, supportsPromptCache: true, supportsVerbosity: true, supportsTemperature: false
- reasoningEffort: high (Responses API only)
- pricing: $15/1M input tokens, $120/1M output tokens
- Set disableTimeout: true to avoid requiring a global timeout override
- Description clarifies: this is a slow, reasoning‑focused model designed for tough problems; requests may take several minutes; it does not stream (UI may appear idle until completion)
- Introduce ModelInfo.disableTimeout to opt out of request timeouts on a per-model basis
- Apply in OpenAI-compatible, Ollama, and LM Studio providers (timeout=0 when flag is true)
- Preserve global “API Request Timeout” behavior (0 still disables globally); per-model flag takes precedence for that model
- Motivation: gpt-5-pro often requires longer runtimes; per-model override avoids forcing a global setting that impacts all models
- Add/extend unit tests to validate provider behavior
* Add a way to save screenshots from the browser tool
* fix: use cross-platform paths in BrowserSession screenshot tests
* fix: validate screenshot paths to prevent filesystem escape
---------
Co-authored-by: Roo Code <roomote@roocode.com>
Adds a new search_replace tool that performs a single search and replace
operation on a file, requiring the old_string to uniquely identify the
target text with 3-5 lines of context.
Parameters:
- file_path: Path to file (relative or absolute)
- old_string: Text to find (must be unique in file)
- new_string: Replacement text (must differ from old_string)
* fix: add Kimi, MiniMax, and Qwen model configurations for Bedrock
- Add moonshot.kimi-k2-thinking with 32K max tokens and 256K context
- Add minimax.minimax-m2 with 16K max tokens and 230K context
- Add qwen.qwen3-next-80b-a3b with 8K max tokens and 262K context
- Add qwen.qwen3-coder-480b-a35b-v1:0 with 8K max tokens and 262K context
All models configured with native tool support and appropriate pricing.
Fixes#9902
* fix: add preserveReasoning flag and update Kimi K2 context window
- Added preserveReasoning: true to moonshot.kimi-k2-thinking model
- Added preserveReasoning: true to minimax.minimax-m2 model
- Updated Kimi K2 context window from 256_000 to 262_144
These changes ensure:
1. Reasoning traces are properly preserved for both models
2. Roo correctly recognizes task completion
3. Tool calls within reasoning traces are handled appropriately
4. Context window matches AWS Console specification
* fix: update MiniMax M2 context window to 196_608 for Bedrock
Based on AWS CLI testing, the actual context window limit for MiniMax M2
on Bedrock is 196,608 tokens, not 230,000 as initially configured.
* Update packages/types/src/providers/bedrock.ts
Co-authored-by: roomote[bot] <219738659+roomote[bot]@users.noreply.github.com>
---------
Co-authored-by: Roo Code <roomote@roocode.com>
Co-authored-by: Matt Rubens <mrubens@users.noreply.github.com>
Co-authored-by: roomote[bot] <219738659+roomote[bot]@users.noreply.github.com>
