heymrun/heym

What it solves

Heym is a self-hosted platform designed to move agentic systems from simple prototypes to production-ready software. It solves the difficulty of orchestrating complex AI agents, managing their execution state, monitoring costs and performance, and maintaining human-in-the-loop control over autonomous processes.

How it works

Heym provides a runtime and a visual canvas for building agentic workflows. Users can compose agents, deterministic logic (loops, switches, merges), and tools into a graph. This graph is then executed by a runtime that manages data flow, parallel execution, and error recovery. The system integrates with LLMs, RAG (via Qdrant or pgvector), and the Model Context Protocol (MCP) to connect to external tools and data.

Who it’s for

It is built for developers and teams who need to deploy agentic systems on their own infrastructure to maintain control over data, security, and execution, and who require professional-grade observability and evaluation tools.

Highlights

• Visual & AI-Assisted Authoring: Create workflows visually on a canvas or generate them using natural language prompts.
• Human-in-the-Loop: Implement review checkpoints that pause agent execution for human approval, edits, or refusal.
• Comprehensive Observability: Track live execution, model costs, latency, and token usage with OpenTelemetry export support.
• Agentic Kanban: A persistent job system where moving cards between columns triggers specific workflow chains.
• Multi-Agent Orchestration: Support for orchestrator agents that can call independent sub-agents and sub-workflows concurrently.
• Extensible Tooling: Integration with Playwright for browser automation and Codex/OpenCode Go for repository-level coding tasks.