convex-agents

You can build stateful AI agents with real-time streaming by using the Convex Agent component to maintain conversation threads, execute tools, and orchestrate workflows alongside an LLM provider like OpenAI.

You can implement Retrieval Augmented Generation (RAG) patterns for AI agents using Convex to retrieve knowledge and deliver context-aware responses while maintaining persistent conversation history.

Yes, integrating an LLM provider like OpenAI is required for workflow orchestration with Convex agents. Convex provides backend durability and state management, while the LLM provider handles conversational reasoning.

AI agents execute external tools and Convex mutations through the Convex Agent component, which allows agents to call external functions and Convex queries directly as tools during multi-step agentic workflows.

The best way to manage conversation history in AI agents is using thread management. The Convex Agent component maintains persistent context and conversation history across user interactions for stateful applications.

Limitations when building multi-step agentic workflows on Convex involve dependency constraints, as the framework explicitly requires the @convex-dev/agent, ai, and openai packages to handle state management and tool execution.