AI Agents, built on Convex. This project provides the Convex Agent Component, a framework for building AI agents with persistent chat history. It enables developers to separate complex agentic workflows from UIs, maintaining reactivity and efficient real-time updates for sophisticated AI interactions.

How It Works

The Agent component abstracts LLM usage, managing models, prompts, and tool calls within agents cooperating in defined workflows. It features persistent message threads with automatic conversation context, hybrid vector/text search, and optional cross-thread search. Responses stream efficiently via WebSockets using deltas. RAG techniques enhance prompts, and durable multi-step workflows, file handling, and debugging tools like an agent playground are included.

Quick Start & Requirements

Install via npm: npm i @convex-dev/agent. Clone the repo (git clone https://github.com/get-convex/agent.git), navigate (cd agent), and run setup/dev commands (npm run setup, npm run dev). Requires Node.js and the Convex platform. Documentation and example links are referenced.

Highlighted Details

• Persistent Chat History: Durable threads shared across users/agents.
• Efficient Streaming: Deltas over WebSockets for real-time updates.
• Contextual Awareness: Auto-includes history, hybrid search (vector/text), opt-in cross-thread search.
• RAG Integration: Supports Retrieval Augmented Generation for prompt context.
• Durable Workflows: Reliable, multi-step agent operations.
• Debugging: Agent playground and dashboard inspection.
• Usage Control: Built-in usage tracking and rate limiting.

Maintenance & Community

Bug reports and feature requests are handled via GitHub issues. Specific community channels or roadmaps are not detailed in the provided project description.

Licensing & Compatibility

License type and compatibility notes for commercial use or closed-source linking are not detailed in the provided project description.

Limitations & Caveats

Specific limitations, unsupported platforms, or known bugs are not detailed. Adoption is contingent on using the Convex platform.