@convex-dev/agent

The @convex-dev/agent component is a core building block for building AI agents on Convex. It manages threads and messages, around which agents can cooperate in static or dynamic workflows. Agents organize LLM prompting with associated models, prompts, and tools, and can generate and stream both text and objects. They can be used in any Convex action, allowing agentic code to be written alongside other business logic. Threads persist messages and can be shared by multiple users and agents, including human agents. Conversation context is automatically included in each LLM call, including built-in hybrid vector/text search for messages. Advanced features include workflows for multi-step operations that span agents and users durably and reliably, RAG techniques for prompt augmentation, and files in chat history with automatic saving to file storage. Debugging is supported via an agent playground to inspect metadata and iterate on prompts and context settings. Usage tracking enables usage billing for users and teams, and rate limiting helps control interaction rates and avoid exceeding LLM provider limits.