Projects

Open source built on one conviction: you should be able to read, debug, and own every part of your AI stack.

Atomic Agents

Building AI agents, atomically.

A lightweight, modular framework for building AI agents in plain Python, built on Instructor and Pydantic for structured outputs. Control flow stays in your code: no hidden LLM calls, no opaque abstractions. Each agent, tool, and context provider is small and single-purpose, the way design-system components are.

View on GitHub
GitHub stars
~6,000
PyPI downloads per month, as of June 2026
20,000+
forks
500+
license
MIT

Tesseron

Expose typed web-app actions to MCP-compatible AI agents.

MCP reaches your systems of record. Tesseron reaches live applications: it lets AI agents call the same handlers your UI already calls, over WebSocket, with typed schemas. Handlers can talk back to agents to confirm, report progress, or ask for structured input. TypeScript SDK plus a matching MCP gateway. v1.0 shipped in April 2026.

View on GitHub
shipped April 2026
v1.0
license
Business Source License 1.1