fix: anthropic models fail when both topP and temp is passed

[felarof99]

No description provided.

[felarof99]

I don't think this is the right fix. @shivammittal274 shouldn't vercel AI SDK handle these for us, could you take look?

[greptile-apps]

Greptile Overview

Greptile Summary

Fixed a compatibility issue where Anthropic models fail when both topP and temperature parameters are passed. The fix introduces a getSamplingParams() method that conditionally omits topP for Anthropic and BrowserOS providers while keeping it for other providers.

Key changes:

• Added provider field to track which AI provider is being used
• Created getSamplingParams() helper that returns only temperature for @anthropic and @browseros, but both temperature and topP for other providers
• Applied the conditional logic to both generateContent() and generateContentStream() methods

Suggestions:

• Consider adding unit tests to verify the parameter filtering works correctly for different providers

Confidence Score: 4/5

• This PR is safe to merge with minimal risk
• The fix is straightforward and addresses a real API compatibility issue. The implementation is clean and follows the existing code patterns. Score is 4 instead of 5 due to lack of unit tests for the new functionality.
• No files require special attention

Important Files Changed

File Analysis

Filename	Score	Overview
packages/agent/src/agent/gemini-vercel-sdk-adapter/index.ts	4/5	Added provider-specific sampling parameter handling to fix Anthropic compatibility issue. Implementation is clean but lacks tests.

Sequence Diagram

sequenceDiagram
    participant Client
    participant ContentGenerator
    participant getSamplingParams
    participant VercelAI
    participant Provider

    Client->>ContentGenerator: generateContent(request)
    ContentGenerator->>getSamplingParams: getSamplingParams(temperature, topP)
    
    alt Provider is Anthropic or BrowserOS
        getSamplingParams-->>ContentGenerator: {temperature}
    else Other providers
        getSamplingParams-->>ContentGenerator: {temperature, topP}
    end
    
    ContentGenerator->>VercelAI: generateText({model, messages, ...samplingParams})
    VercelAI->>Provider: API call with filtered params
    Provider-->>VercelAI: response
    VercelAI-->>ContentGenerator: result
    ContentGenerator-->>Client: GenerateContentResponse

Loading

[greptile-apps]

_{1 file reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}