Open
Conversation
Replace OpenAILLM inheritance with native Google GenAI SDK implementation to fix 400 errors caused by unsupported parameters (frequency_penalty, presence_penalty) and incompatible APIs. Changes: - Implement BaseLLM<GeminiConfig> directly instead of extending OpenAILLM - Add message conversion for Gemini format (system→systemInstruction, assistant→model, tool→functionResponse) - Add tool declaration conversion using z.toJSONSchema() - Implement all required methods: streamText, generateText, generateObject, streamObject - Remove baseURL parameter from provider (no longer using OpenAI compat layer) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Remove Date.now() from generateToolCallId hash input to ensure consistent IDs across streaming chunks. This prevents duplicate entries in accumulatedToolCalls when the same tool call appears in multiple chunks. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add validation before accessing response properties to provide clear error messages when API returns empty/blocked responses. Includes finish reason inspection for better diagnostics. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Replace custom inline options type with GenerateOptions from shared types module for consistency with other providers. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add additionalInfo with finishReason to generateText and streamText - Track lastFinishReason during streaming to include in final yield - Add comment explaining 0.7 default temperature (vs OpenAI's 1.0) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fix tool call ID collision by using monotonic counter instead of chunk-local index in streamText - Remove misleading empty object yield on parse errors in streamObject - Add final validation after stream ends to ensure valid JSON output Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Gemini 3 models require thoughtSignature to be preserved and sent back during function calling. This adds: - thoughtSignature field to ToolCall type - Extract thoughtSignature from Gemini functionCall responses - Include thoughtSignature when converting messages back to Gemini format This fixes "Function call is missing a thought_signature" error with gemini-3-pro-preview and gemini-3-flash-preview models. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Author
|
@codex 이거 PR 리뷰 좀 부탁해요! |
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
When updating existing tool calls during streaming, only arguments was being copied, losing the thoughtSignature required by Gemini 3 models in multi-turn conversations. Fixes: "Function call is missing a thought_signature" error in Researcher agent with gemini-3-*-preview models. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The Gemini API returns model names with "models/" prefix (e.g., "models/gemini-3-flash-preview"), but the @google/genai SDK adds this prefix internally. This caused duplicate prefix "models/models/..." resulting in 400 Bad Request errors. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
thoughtSignature is a sibling of functionCall in the Part object, not a property of FunctionCall. Fixed both generateText and streamText to access raw parts from candidates[0].content.parts to correctly extract thoughtSignature. Before: (fc as any).thoughtSignature // Wrong: fc is FunctionCall After: (part as any).thoughtSignature // Correct: part contains both Fixes "Function call is missing a thought_signature" error with Gemini 3 models in multi-turn tool calling conversations. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary by cubic
Rewrite Gemini provider to use the native @google/genai SDK instead of the OpenAI-compat layer. Fixes 400 errors from unsupported params and adds full tool-call and streaming support.
New Features
Bug Fixes
Written for commit 856c1a8. Summary will update on new commits.