Heym

// Overview

Heym is a self-hosted runtime and platform for building, orchestrating, running, observing, evaluating and controlling agentic systems on your own infrastructure. In Heym a workflow is a graph that defines how agents, deterministic steps, tools and data work together; the runtime executes that graph, runs independent branches in parallel, records execution history and supports human review checkpoints. The project stresses that the visual canvas is only one interface into the runtime — the same workflows also run through REST and SSE APIs, cron schedules, event triggers, MCP clients and Portal chat pages.

Workflows are composed from a node library that mixes LLM, Agent, RAG, Codex and OpenCode Go nodes with conditions, switches, loops, merges, data transformations and sub-workflow calls, or generated from a plain-text or voice description by the built-in AI Assistant. An orchestrator agent can delegate to named sub-agents and sub-workflows and call independent ones concurrently. Agents can retrieve documents from Qdrant or PostgreSQL with pgvector, keep a per-agent knowledge graph as persistent memory, call MCP servers, and load reusable SKILL.md skills with optional Python tools. Built-in nodes cover Slack, Discord, Telegram, email, GitHub, Jira, Linear, Notion, Google Sheets, BigQuery, Redis, RabbitMQ, S3, Playwright browser automation and more.

The operations side is where Heym puts its weight: execution history with live attach to running jobs, LLM traces with tokens, latency and USD cost per call, analytics, alerts on errors, duration or spend, prompt evals across models, input guardrails, OIDC single sign-on, team sharing and OpenTelemetry export. The backend is Python with FastAPI and PostgreSQL, the frontend Vue 3 with TypeScript. Heym is source-available under the MIT License with the Commons Clause condition: you may use, modify and self-host it, but selling a product or service whose value derives substantially from Heym requires a commercial licence from the maintainers.

// What it does

• Visual canvas plus an AI Assistant that generates nodes and edges from a text or voice description, with reusable templates
• Multi-agent orchestration: an orchestrator agent delegates to named sub-agents and sub-workflows, running independent calls concurrently
• Built-in RAG over Qdrant or PostgreSQL/pgvector, per-agent persistent memory as a knowledge graph, and SKILL.md agent skills with optional Python tools
• MCP both ways — agents call MCP servers over SSE, Streamable HTTP or stdio, and workflows are published as MCP tools at /api/mcp/sse
• Human-in-the-loop checkpoints that pause a run for Accept, Edit & Continue or Refuse and resume from a saved snapshot
• Observability built in: execution history, LLM traces with token and USD cost, analytics, alerts, prompt evals across models and OpenTelemetry export

// Getting started

Heym needs PostgreSQL as its core database. Locally, run.sh starts PostgreSQL, the backend and the frontend for you; for servers there is a Docker Compose deployer and a prebuilt image. Prerequisites for the local path are Bun, Python 3.11+, uv and Docker.

git clone https://github.com/heymrun/heym.git
cd heym
./run.sh

cp .env.example .env
./deploy.sh              # Build and deploy
./deploy.sh --logs       # View logs
./deploy.sh --down       # Stop services

POST /api/workflows/{workflow_id}/execute         # REST
POST /api/workflows/{workflow_id}/execute/stream  # SSE streaming
/api/mcp/sse                                      # MCP server endpoint

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Run multi-agent workflows on your own infrastructure instead of a hosted automation SaaS, with credentials encrypted and access gated by SSO
• Put a human approval step in front of agent actions — a reviewer accepts, edits or refuses, and the run resumes from its saved state
• Expose an internal workflow as an MCP tool for Claude Desktop or Cursor, or as a chat Portal for colleagues, without duplicating the logic
• Track what agent workflows cost and how they fail — per-call traces, USD spend by model, alerts on errors or spend, and OpenTelemetry spans for your existing tracing backend

// How Heym compares

Heym alongside other open-source workflow automation tools AI/TLDR tracks, ranked by GitHub stars.