Open Source AI Agent Framework

Kubernetes for
AI Agents

Self-hosted. Production-grade. TypeScript-native.

View on GitHubProduct Hunt
cogitator — bash

Everything you need for production AI

From local development to global scale. No vendor lock-in.

Multi-Model Runtime

Run Ollama, vLLM, OpenAI, Anthropic, or Google models with identical code. Switch providers without changing a line.

DAG Workflows

Build complex agent pipelines with retry, compensation, and human-in-the-loop steps.

Multi-Agent Swarms

6 coordination strategies out of the box. Hierarchical, consensus, round-robin, and more.

Production Memory

Redis for speed, Postgres for persistence, pgvector for semantic search. Your agents remember everything.

Sandboxed Execution

Run untrusted code in Docker containers or WASM. Never on your host.

MCP Protocol

First-class support for Model Context Protocol. Connect any MCP server as a tool.

Server Adapters

Express, Fastify, Hono, Koa — mount agents as REST APIs with SSE streaming, WebSocket, and auto-generated Swagger docs.

Full Observability

OpenTelemetry traces, Prometheus metrics, cost tracking. Know exactly what your agents are doing.

Cogitator

Built for engineers who trust their agents

GitHubDocsCookbookDiscord
© 2026 • MIT License