You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The framework provides a unified interface for integrating with various LLM providers, including OpenAI, Anthropic, Google, Groq, Cohere, Mistral, Bedrock, and HuggingFace. Each model integration follows a consistent settings pattern with provider-specific prefixes (e.g., `google_*`, `anthropic_*`). \n\
4928
4928
4929
4929
Examples of supported models and their capabilities include:
4930
-
* `GoogleModel`: Integrates with Google's Gemini API, supporting both Gemini API (`google-gla`) and Vertex AI (`google-vertex`) providers. It supports token counting, streaming, built-in tools like `WebSearchTool`, `WebFetchTool`, `CodeExecutionTool`, and native JSON schema output. \n\
4930
+
* `GoogleModel`: Integrates with Google's Gemini API, supporting both Gemini API (`google-gla`) and Vertex AI (`google-vertex`) providers. It supports token counting, streaming, built-in tools like `WebSearchTool`, `UrlContextTool`, `CodeExecutionTool`, and native JSON schema output. \n\
4931
4931
* `AnthropicModel`: Uses Anthropic's beta API for advanced features like "Thinking Blocks" and built-in tools. \n\
4932
4932
* `GroqModel`: Offers high-speed inference and specialized reasoning support with configurable reasoning formats. \n\
Tools can return various types of output, including anything Pydantic can serialize to JSON, as well as multimodal content like `AudioUrl`, `VideoUrl`, `ImageUrl`, or `DocumentUrl`. The `ToolReturn` object allows for separating the `return_value` (for the model), `content` (for additional context), and `metadata` (for application-specific use). \n\
4946
4946
4947
-
Built-in tools like `WebFetchTool` allow agents to pull web content into their context. \n\
4947
+
Built-in tools like `UrlContextTool` allow agents to pull web content into their context. \n\
0 commit comments