Prashant lib

[prashant4654]

This pull request introduces several improvements and fixes related to Google GenAI integration, especially around handling "reasoning" (thought) content in LLM responses, configuration, and example usage. The most significant changes are the improved extraction and processing of reasoning content, updates to model/provider handling, and enhancements to example scripts for better developer experience.

Reasoning Content Extraction & Processing:

• Improved extraction of reasoning tokens by using thoughts_token_count from the response metadata instead of hardcoding to zero.
• Refined the logic for processing parts in LLM responses: reasoning parts are now handled before text parts, ensuring that reasoning content is not duplicated or missed. [1] [2] [3]

Model/Provider Handling & Configuration:

• Fixed model/provider splitting logic in Agent initialization to properly assign the model name.
• When building the Google config, added a thinking_config with include_thoughts=True for text output, enabling the model to return reasoning content in its response.
• Simplified Google GenAI client creation by removing unnecessary keyword arguments.

Example Script Improvements:

• Updated the example in graph.py to load environment variables using dotenv and ensure GEMINI_API_KEY is set correctly.
• Changed the model string from gemini/gemini-2.5-flash to google/gemini-2.5-flash for consistency with provider/model conventions.
• Modified the example input message to ask about the weather in London, making the example more relevant.

[Copilot]

Pull request overview

This pull request improves Google GenAI integration in the Agentflow framework, focusing on proper handling of reasoning (thought) content in LLM responses, fixing model/provider splitting logic, and updating configuration. The changes enable the framework to correctly extract and process reasoning tokens from Google's Gemini models that support thinking/reasoning capabilities.

Changes:

• Fixed critical bug in Agent model/provider splitting logic that prevented proper model name extraction
• Enhanced reasoning content extraction to handle Google GenAI's thought parts where part.thought=True with content in part.text
• Added ThinkingConfig to Google GenAI requests to enable reasoning content in responses
• Updated example to use correct google/ provider prefix and improved dotenv handling

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
examples/agent-class/graph.py	Updated model prefix from gemini/ to google/ for Agent class consistency; improved environment variable loading (with redundant logic); changed example query to weather-related
agentflow/graph/agent.py	Fixed model name assignment after provider/model split; removed invalid llm_kwargs from Google client initialization; added ThinkingConfig to enable reasoning content extraction
agentflow/adapters/llm/google_genai_converter.py	Enhanced reasoning token extraction from metadata; refactored part processing order to prevent content duplication; updated reasoning part handling to correctly extract content from part.text when part.thought is True

[codecov]

Codecov Report

❌ Patch coverage is 51.33690% with 182 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...entflow/adapters/llm/openai_responses_converter.py	51.17%	87 Missing and 17 partials ⚠️
agentflow/graph/agent.py	39.50%	38 Missing and 11 partials ⚠️
agentflow/adapters/llm/openai_converter.py	45.71%	18 Missing and 1 partial ⚠️
agentflow/adapters/llm/google_genai_converter.py	38.46%	5 Missing and 3 partials ⚠️
agentflow/adapters/llm/reasoning_utils.py	92.30%	1 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

[Copilot]

Pull request overview

Copilot reviewed 17 out of 17 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

agentflow/adapters/llm/openai_converter.py:144

• The order of content blocks appears inconsistent with the Google converter. In the Google converter, ReasoningBlocks are added before TextBlocks (see google_genai_converter.py lines 141-146), but here TextBlocks are added before ReasoningBlocks. For consistency with the PR's stated goal of "processing reasoning parts before text parts", consider reordering these blocks to add ReasoningBlock first, then TextBlock.

        blocks = []
        if content:
            blocks.append(TextBlock(text=content))
        if reasoning_content:
            blocks.append(ReasoningBlock(summary=reasoning_content))