- Switch from the sidebar to the call method to see how to invoke Agent Party through OpenAI API, MCP server, docker, and the web interface. The OpenAI interface has added the following switch parameters:
- enable_thinking: The default is False, whether to enable the thinking mode.
- enable_deep_research: The default is False, whether to enable the deep research mode.
- enable_web_search: The default is False, whether to enable web search.
- Knowledge base, allowing the large model to answer questions based on the information in the knowledge base. And it supports the following functions:
- If there are multiple knowledge bases, the model will actively query the corresponding knowledge base according to the question requirements.
- You can choose the timing of retrieval, and you can choose to actively retrieve or passively retrieve the knowledge base.
- We have supported the rerank model, which can improve the retrieval effect of the knowledge base.
- Support mixed search function, which allows you to choose the proportion between keyword search and semantic search.
- Networking function, which allows the large model to actively query information online according to the needs of the question. Currently, it supports:
- Search Engine
- duckduckgo (Completely free, but cannot be accessed in China's online environment)
- searxng (can be locally deployed with Docker)
- tavily（需要申请api key）
- bing (need to apply for an API key)
- google (need to apply for an API key)
- Brave (need to apply for an api key)
- exa (need to apply for an API key)
- serper (need to apply for an API key)
- bochaai (need to apply for an API key)