Heym — self-hosted agent platform

★ 1.3K

Heym is an open-source, self-hosted platform and runtime for building, orchestrating, running, observing, evaluating and controlling agentic systems on your own infrastructure. Reach for it when you need AI agents with human approval checkpoints before sensitive actions, when data cannot leave your servers, when you want one environment for both visual and code-defined workflows, or when you need to track per-run token cost and latency. Inside: a visual canvas and an AI assistant that generates workflows; agents and deterministic nodes, sub-agents, parallel branches, retries and model fallback; RAG on Qdrant or PostgreSQL with pgvector; integrations via MCP; Playwright browser automation; run history, traces, model costs and OpenTelemetry export; prompt evaluation and cross-model comparison; and a Kanban board of agentic jobs. Workflows are invoked through REST, SSE, schedules, events, MCP and Portal chat. The backend uses FastAPI, the frontend Vue 3 and TypeScript. Try it locally with git clone followed by ./run.sh. The repository includes a COMMONS-CLAUSE.md file, so check the licence terms before commercial use. It orchestrates agents on your own servers; it is not a hosted no-code service.