Build, execute, govern, observe, and evaluate AI agents on one self-hosted runtime. Heym is an AI workflow platform you run on your own infrastructure.
Product tour
See the problems Heym removes, how each capability works, and why it matters in practice.
Overview
The problem
Teams keep stitching together separate tools for work that should feel simple.
The Heym answer
Heym brings sixteen once-difficult jobs onto one visual canvas.
Use the arrow keys to move through the tour
Why it matters
This tour starts with the problem, then shows the Heym answer. Each stop focuses on a real task teams already handle today, without burying the point under a long feature list.
Heym product demo: a drag-and-drop canvas where agentic AI workflow nodes such as LLM, Agent, RAG / Vector Store, HTTP, and Slack are connected into a running automation pipeline.
A prototype becomes a production system when five things hold at once. Heym is an AI workflow platform that carries all five in one runtime you host yourself, so an agent that works in a demo keeps working when it runs a thousand times a day.
Lay out agents, tools, and logic on a visual canvas, or describe the workflow in plain language and let the AI assistant draw it for you.
Run those workflows at production scale, with branches that execute in parallel, retries on failure, and worker instances that share the load.
Decide what an agent may do on its own and what waits for a person, then keep a record of every privileged action.
See what every model call and tool call did on every run, how long it took, and what it cost, then get told when one of those numbers crosses a line.
Score prompts, models, and agent output against a repeatable suite so a change is measured before it reaches production.
The same runtime covers all of it. Self-hosted AI agents run on your own infrastructure with Docker or Kubernetes, your credentials and run data stay inside your environment, and the source is available under an MIT license with a Commons Clause.
Heym focuses on ten solution areas where controlled, observable AI operations matter most. Each one is backed by real workflows you can import and run.
Grounded employee and customer answers from corporate sources.
Sensitive workflows with human validation on key actions.
Specialized agents to collect, evaluate, and report data.
AI that auto-repairs broken browser selectors.
Collaborative reusable AI skills with execution history.
Real-time agent activity logs and actionable charts.
Trace every LLM token cost before scaling up.
Deploy logic flows as API endpoints or MCP servers.
Performance metrics before shipping code.
Trigger Slack, email, or webhook alerts on thresholds.
Every component is chosen for performance, developer experience, and reliability.
The frontend uses Vue.js 3 with TypeScript for a fast, type-safe development experience. Vite handles builds and Tailwind CSS takes care of styling.
A Python backend powered by FastAPI delivers async performance for concurrent workflow executions. Alembic handles database schema migrations.
PostgreSQL stores workflows and execution history, while pgvector powers built-in vector search for RAG pipelines. Docker keeps local development consistent.
JWT tokens in HttpOnly cookies with refresh token rotation secure user sessions. Passwords use bcrypt hashing, credentials are encrypted at rest with Fernet.
Connect to OpenAI, Ollama for local models, vLLM for high-throughput inference, or expose workflows through the Model Context Protocol.
First-party documentation lives inside Heym, alongside a native chat-with-docs experience. ESLint and Ruff keep the codebase healthy.
Answers to your common questions about Heym.
Still have questions?
Ask on Discord