Heym: The Free, Self-Hosted AI Workflow Automation Platform With Multi-Agent Canvas and MCP Support (2026 Guide)

Legacy automation platforms like Zapier and Make were built for a world of simple IF/THEN rules. They connect APIs together, trigger on webhooks, and move data from point A to point B. But when you try to build complex, long-running AI workflows with multi-agent loops, RAG pipelines, and human-in-the-loop approvals on those platforms, you end up fighting the UI more than building your product. AI was bolted on as an afterthought, and it shows.

Heym is a free, self-hosted, AI-native workflow automation platform built from the ground up around LLMs, agents, and intelligent tooling. It gives you a visual canvas where you can wire together multi-agent orchestrations, RAG pipelines, web scrapers, and HTTP calls in a single unified runtime. It supports the Model Context Protocol (MCP) out of the box, runs entirely in Docker, and gives you deep observability into every LLM call including token costs and latency.

In this guide, you'll learn what Heym is, who it's for, and how to deploy it with Docker and build your first AI workflow on the visual canvas.

What is Heym?

Heym is an open source, self-hosted workflow automation platform designed specifically for the AI era. While traditional automation tools were built for rule-based logic and had AI capabilities added later, Heym is purpose-built around LLMs, autonomous agents, and intelligent tooling from the ground up. The entire platform runs as a unified runtime, so you do not need to glue together five different services to handle your agents, document retrieval, and monitoring.

The core of Heym is a visual canvas where you drag and drop nodes to build multi-agent orchestrations, RAG pipelines, web scrapers, and HTTP integrations. You can wire agents to run in parallel or sequentially, chain their outputs, and add human-in-the-Loop approval gates that pause the workflow until a human signs off. This is critical for workflows that involve sensitive actions like sending emails or spending money, where you want a human to review the agent's decision before it executes.

Heym also supports the Model Context Protocol (MCP) natively, which means you can attach external skills and tools to your agents instantly, or expose your entire Heym workflow as a tool for another AI to call. Because it is entirely self-hosted via Docker, you maintain absolute control over your data with zero vendor lock-in. The platform includes granular LLM tracing that shows you the exact request and response payload, latency, and token cost for every single LLM call, not just a green checkmark.

Who is it for?

AI engineers and developers: Developers who are tired of fighting traditional automation platforms when trying to build complex, long-running agentic loops and multi-agent systems.
No-code builders: People who want to create powerful multi-agent systems using a drag-and-drop visual canvas rather than writing and maintaining massive Python scripts.
Operations teams: Teams needing human-in-the-loop approvals. Heym natively pauses workflows for human sign-off before executing sensitive actions like sending an email or spending money.
Data-driven teams: Teams who require deep observability into their AI workflows. You need to know exactly how much an agent costs per run and why it produced a specific output, and Heym provides this natively.

What makes Heym different from Zapier and Make?

Generate workflows by prompt: You do not even have to drag and drop. Using the built-in AI Assistant, you can type "Create a workflow with a Research Agent and a Writer Agent running in parallel, orchestrated by a Manager" and Heym will draw the complete graph for you automatically.
AI Convert migration tool: Have an existing automation stuck in another platform like n8n or Zapier? Heym's AI Convert tool can analyze your old workflow JSON and automatically translate the nodes and connections into a native Heym canvas.
Granular LLM tracing: Unlike basic workflow tools that just show a green checkmark, Heym traces every individual LLM call. You can click any node to see the exact request and response payload, latency, and token cost in real time.
Reusable skills: You can generate agent skills from natural language, preview the generated SKILL.md file, and attach them across multiple different workflows without rewriting any code.
Native MCP support: Built-in Model Context Protocol server and client capabilities let you attach external tools to your agents or expose your workflows as callable tools for other AI systems.
Free and open source: The entire platform is free and source-available. You pay only for your own hosting and LLM API keys, with no per-task pricing or vendor lock-in.

What you need before you start

Heym is designed to be self-hosted on your own infrastructure to protect your proprietary data. Make sure you have the following:

Docker and Docker Compose: The standard and easiest way to deploy the platform locally or on a cloud server like DigitalOcean, AWS, or Hetzner.
LLM API keys: API credentials for the models you intend to use. Heym supports OpenAI, Anthropic, Gemini, and local providers via OpenAI-compatible endpoints.
A modern web browser: To access the visual canvas UI once the platform is running.
Git: To clone the repository from GitHub.
A server or local machine: At least 2 GB RAM for a local test deployment. Production deployments with multiple concurrent workflows will benefit from 4 GB or more.

Step-by-step installation

Step 1: Clone the repository

Open your terminal and clone the official Heym GitHub repository to your local machine or server:

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

Step 2: Configure environment variables

Copy the example environment file to create your own configuration:

cp .env.example .env

Open the .env file in your text editor. Here you can configure your database settings, default ports, and the API keys for the LLM providers you plan to use. At minimum, add your OpenAI or Anthropic API key so the orchestrator agent can function.

Step 3: Deploy via Docker Compose

Once your environment is configured, use Docker Compose to pull the necessary images and spin up the platform. This includes the backend API, the frontend canvas, and the database:

docker compose up -d

The first run will take a few minutes as Docker pulls the images and initializes the database. You can check the status of all containers with docker compose ps.

Step 4: Access the canvas

Once the containers are running, open your web browser and navigate to http://localhost:3000 or whichever port you defined in your .env file. You will see the Heym visual canvas.

You can instantly begin clicking the canvas to add nodes manually, or use the AI Assistant chat box to describe your desired automation in natural language and watch the nodes assemble themselves. Try a prompt like "Create a workflow that scrapes a URL, summarizes the content with an LLM, and sends the summary to a Slack webhook" to see the graph generation in action.

Common errors and how to fix them

Error	What it means	How to fix it
Docker port conflict (port 3000 or 5432 already in use)	You are already running a web server, Next.js app, or PostgreSQL database on your machine that is blocking Heym from binding to its default ports.	Open your `docker-compose.yml` or `.env` file and map Heym's external ports to something else, for example change `3000:3000` to `3005:3000` and `5432:5432` to `5433:5432`.
Long-running agent workflows silently fail or timeout	Complex multi-agent tasks like heavy web scraping and summarization can exceed standard HTTP request timeouts or hit LLM API rate limits.	Open the Traces tab in Heym. It shows the exact node where the failure occurred. If it is a timeout, adjust your container timeout settings. If it is a rate limit, the trace will display the exact 429 error from your LLM provider.
AI Convert fails to migrate an old workflow	You pasted JSON from a legacy platform like Zapier or Make, but it contains proprietary closed-source app integrations that Heym cannot perfectly map.	AI Convert handles logical flow, HTTP requests, and data routing well, but you may need to manually reconfigure specific third-party SaaS authentication nodes like custom Salesforce or Slack OAuth connections on the canvas after the initial conversion.
Container exits immediately after startup	The database container failed to initialize, often due to insufficient disk space or a corrupted volume from a previous failed deployment.	Run `docker compose down -v` to remove the old volumes, then `docker compose up -d` to start fresh. Ensure you have at least 5 GB of free disk space.

Heym vs traditional automation platforms

Feature	Heym (self-hosted)	Traditional automation (Zapier, Make)
Cost	Free, pay only for your own hosting and API keys	Expensive monthly subscriptions per task
Agentic orchestration	Native, designed for autonomous agents and RAG	Rigid, designed for static IF/THEN rules
LLM observability	Deep, node-level traces with token costs and latency	Shallow, usually just pass or fail logs
Model Context Protocol	Yes, built-in server and client capabilities	No
Data sovereignty	Full control, runs on your own infrastructure	Data routes through vendor cloud servers
Best for	Teams building complex AI-native workflows	Teams connecting legacy SaaS apps with simple triggers

Bottom line: If you are trying to build complex AI workflows on platforms built five years ago, you are likely fighting the UI more than building your product. Heym bakes agents, RAG, human-in-the-loop approvals, and deep observability directly into the visual canvas, eliminating the need to bolt together three different platforms just to get a reliable AI automation running. While Zapier charges per task and limits you to rigid rule-based logic, Heym is free, self-hosted, and built purely for the AI era. The ability to deploy it via Docker and use MCP out of the box makes it the best open source option for privacy-conscious teams building production AI workflows.

3 alternatives worth checking out

n8n (n8n.io): The titan of open source workflow automation. n8n is incredibly mature, boasts hundreds of pre-built app integrations, and has added AI agent and memory support. While Heym is built purely around AI from day one, n8n is the better choice if your workflow relies heavily on connecting to dozens of legacy SaaS apps like Salesforce, Shopify, or QuickBooks.
Flowise (github.com/FlowiseAI/Flowise): An open source drag-and-drop UI explicitly designed for building customized LLM flows and LangChain applications. Flowise is fantastic for rapidly prototyping AI chatbots and RAG systems visually, though Heym offers a more robust environment for broader workflow automation and multi-agent orchestration beyond just chat interfaces.
Dify (github.com/langgenius/dify): An open source LLM application development platform that combines AI workflow orchestration with deep RAG capabilities and a polished prompt-engineering dashboard. It sits between Flowise for chatbot building and Heym for workflow automation, making it a strong alternative for enterprise teams looking to deploy production AI apps.

Found this guide useful? Check out more AI tools and open source projects on Sudo Scout.