heymrun/heym
Build AI workflows by prompt or visual canvas. Heym is source-available and self-hosted, with agents, RAG, MCP, HITL, observability, evals, token cost tracking and more.
What it solves
Heym is an AI-native automation platform that allows users to build, visualize, and run intelligent workflows without writing code. It addresses the limitation of traditional automation tools that treat AI as an add-on, instead making AI the core execution model for complex tasks like multi-agent orchestration and RAG pipelines.
How it works
Users create workflows on a drag-and-drop visual canvas using a library of nodes. These workflows can be generated automatically from natural language or voice prompts via an AI Assistant. The platform supports multi-agent orchestration where an orchestrator agent can trigger sub-agents and sub-workflows in parallel to increase speed and focus. It integrates with vector stores (Qdrant or Postgres/pgvector) for RAG, supports the Model Context Protocol (MCP) for connecting to external servers, and provides a public chat interface (Portal) for end-users.
Who it’s for
Developers and enterprise teams who want a self-hostable, observable, and scalable way to deploy AI agents and automated workflows into production without the "enterprise gatekeeping" of restricted production features.
Highlights
- AI-Native Canvas: Drag-and-drop editor with an AI Assistant that builds workflows from natural language.
- Multi-Agent Orchestration: Support for parallel execution of sub-agents to reduce context bloat and latency.
- Built-in RAG: Integrated semantic search and managed vector stores.
- Observability: Full LLM traces, OpenTelemetry export, and real-time USD cost tracking per execution.
- Human-in-the-Loop: Ability to pause agent execution for user approval or input.
- MCP Support: Connects to MCP servers as a client or exposes workflows as an MCP server.
- Agentic Kanban: Persistent agentic jobs where moving cards triggers specific workflow chains.
- Auto-Heal: AI-driven automatic detection and fixing of broken Playwright selectors during browser automation.