Heym: self-hosted AI workflow runtime built beyond n8n
n8n is a solid tool. So is Zapier. So is Make. They were built for deterministic, rule-based automation, and they do that job well. The problem is that AI-native workflows are not deterministic or rule-based. The moment you add agents that reason, documents that need retrieval, or approval steps before anything consequential fires, those platforms stop being a foundation and start being something you fight against.
That is the problem the developer of Heym ran into. Their response was to build a new runtime from scratch rather than keep patching a tool that was not designed for the job.
️ What Heym Is
Heym is a self-hosted, source-available AI workflow automation platform. It ships as a Docker Compose stack and runs entirely on your own infrastructure. No data leaves your stack.
The stated design goal is a single runtime for everything an AI workflow needs: agents, document retrieval, human-in-the-loop approval checkpoints, observability, scheduling, and the ability to expose any workflow as a callable tool for external AI assistants.
How the Execution Engine Works
The workflow engine builds a directed acyclic graph from the canvas and runs independent nodes concurrently using a thread pool. Streaming mode emits events as each node completes, so the frontend updates in real time rather than waiting for the full run to finish.
Agent nodes run a full tool-calling loop. They can execute Python tools, connect to external MCP servers, delegate to sub-agents, and call other Heym workflows as tools. When context usage approaches 80% of the model window, the engine automatically compresses history. Long-running agents don’t silently fail mid-task because the context filled up.
Human-in-the-Loop as a First-Class Primitive
This is where Heym makes the strongest case for itself. Most automation tools treat human review as an afterthought, a webhook you fire and then poll. Heym treats it as a design primitive.
The HITL node pauses execution at any point in the workflow, generates a public one-time review URL, and waits. A reviewer can accept, edit, or refuse the output without needing a Heym account. When they respond, execution resumes from an exact stored snapshot of where it paused. The same run can pause more than once.
For workflows that touch anything with real consequences, drafted emails, generated reports, data transformations feeding downstream systems, this matters. You get the speed of automation with a human gate exactly where you need it.
Built-in Vector Retrieval
Document retrieval in Heym is not a separate service called via API. It is native to the runtime. You upload documents, create vector stores, and wire semantic search directly into the workflow canvas. The entire retrieval pipeline runs inside a single workflow and shows up in a single trace.
The developer’s argument here is straightforward: bolting retrieval on from outside creates two systems to maintain and two places to debug, with no unified trace across either. Native retrieval eliminates that split.
MCP Server Included
Every Heym instance runs a built-in MCP server. Any workflow you build can be exposed as a tool that Claude Desktop, Cursor, or any MCP-compatible client can call directly. Agent nodes can also connect to external MCP servers as tool sources, so capabilities flow in both directions: into Heym from external tools, and out of Heym to external AI assistants.
Observability and Evals
The Traces tab logs every execution automatically. The Evals tab lets you build test suites and run evaluations across multiple models simultaneously with configurable scoring. According to the developer, this is how they debug their own workflows, not a feature added for marketing purposes.
If you can’t inspect what your AI workflow did, why it did it, and when, you can’t trust it in production. Heym’s position is that observability has to be native, not bolted on after the fact.
️ The Stack
- Frontend: Vue 3 with TypeScript and Vue Flow
- Backend: Python and FastAPI with async SQLAlchemy
- Storage: PostgreSQL 16
- Deployment: Docker Compose
Pro Tip
If you are self-hosting anything AI-related and want a single audit trail across agents, retrieval, and approvals, native observability in the runtime is worth more than you might expect. Debugging a three-system trace is a different job entirely than reading one.
Pricing and Status
Heym is currently at v0.0.1. It is actively developed and released under MIT plus Commons Clause, which means source-available rather than fully open source. Free to self-host. The source is on GitHub.
This is early software. If you’re building AI workflows and find yourself writing more glue code than actual workflow logic, it’s worth a look. The architecture decisions, particularly around HITL as a primitive and native retrieval, are the right instincts for where agent-based automation is heading.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
