Giles Odigwe
4b1afd9052
Gemini thought-summary parts (part.thought=True) were dropped in _parse_parts, so reasoning never reached ChatResponse.contents. Emit them as text_reasoning content instead, matching OpenAIResponsesClient. Round-trip is safe: _convert_message_contents never re-emits reasoning text as a Part. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: b8aa906e-1408-40c1-9a45-6deb40dc36f8
Get Started with Microsoft Agent Framework Gemini
Install the provider package:
pip install agent-framework-gemini --pre
Gemini Integration
The Gemini integration enables Microsoft Agent Framework applications to call Google Gemini models with familiar chat abstractions, including streaming, tool/function calling, and structured output.
Structured Output
Gemini structured output can be configured with either a Pydantic model in response_format, a JSON schema mapping in response_format, or a Gemini-specific response_schema. Declarative agents that define outputSchema pass that schema through response_format.
Authentication
The connector supports both google-genai authentication modes.
Gemini Developer API
Obtain an API key from Google AI Studio and set either the package-prefixed or SDK-standard environment variable:
export GEMINI_API_KEY="your-api-key"
# or: export GOOGLE_API_KEY="your-api-key"
export GEMINI_MODEL="gemini-2.5-flash-lite"
# or: export GOOGLE_MODEL="gemini-2.5-flash-lite"
Vertex AI
Set the standard Vertex AI environment variables used by google-genai:
export GOOGLE_GENAI_USE_VERTEXAI=true
export GOOGLE_CLOUD_PROJECT="your-project-id"
export GOOGLE_CLOUD_LOCATION="global"
export GOOGLE_MODEL="gemini-2.5-flash-lite"
Examples
See the Google Gemini samples for runnable end-to-end scripts covering:
- Basic agent with tool calling and streaming
- Extended thinking with
ThinkingConfig - Google Search grounding
- Google Maps grounding
- Built-in code execution