Heym Review (2026): Pricing, Features & Honest Verdict

Heym is a source-available agentic workflow automation platform you self-host, combining a visual low-code canvas with agent nodes, RAG, MCP support, human approval checkpoints and built-in tracing and evals. Best for: platform teams who need agents running on their own infrastructure. Price: free to self-host, no usage limits (free plan: yes). Rating: 7.5/10.

We found Heym on Product Hunt on 10 August 2026, relaunching after an April debut. It sits in the crowded space between n8n and an agent framework, and its pitch is specific enough to be worth taking seriously: build agentic systems visually, run them on your own hardware, and inspect every execution afterwards.

What is Heym?

Heym is a workflow automation platform where the nodes are AI-native rather than AI-flavoured. You describe a workflow in natural language and an assistant builds it, or you wire it yourself from a library spanning seven categories: triggers, AI, logic, data, integrations, automation and utilities. The result runs as a DAG with parallel execution, and you can deploy any workflow as a portal, an API or an MCP tool.

The demo on the homepage is a support triage flow, and it shows the shape well: an inbound trigger, an intent-and-routing agent, a RAG retrieval step against Qdrant, three specialist workers running in parallel across billing, technical and account, then a human approval checkpoint before anything completes. The trace panel reports 3.8 seconds duration, 4 agent calls, 7 tool calls and an eval score of 0.96. Heym was built by Burak Akgün and Ceren Kaya Akgün, and the code lives at github.com/heymrun/heym with around 902 stars.

What Are Heym’s Key Features?

Agent nodes with real tool calling and Python

Agent nodes support tool calling and Python tools, which puts Heym on the correct side of the line between “LLM in a box” and something you can build a system on. Combined with parallel DAG execution, you can fan a request out to several specialist agents and arbitrate their answers. One of the templates does exactly that for code review: a primary reviewer proposes findings and a challenger running a different model disputes them, arbitrated into one calibrated decision.

Human-in-the-loop checkpoints and guardrails

Approval checkpoints are first-class nodes rather than an afterthought, and content guardrails sit alongside them. For anyone putting agents near customers or production systems, this is the feature that decides whether the platform is usable at all. Heym puts the reviewer in the graph where you can see them.

Built-in RAG, MCP client and MCP server

Retrieval runs over Qdrant or built-in Postgres with pgvector, so you are not bolting on a separate vector service to get started. Heym is both an MCP client and an MCP server, meaning it can consume external tools and expose your workflows as tools to other agents. That bidirectional support is still uncommon and it is the difference between a platform that participates in an agent ecosystem and one that sits outside it.

Tracing, cost tracking and evals in the box

Traces, costs, latency and eval scores are part of the runtime rather than a separate observability purchase. Anyone who has run agents in production knows the failure mode: it works in the demo, it costs four times what you expected in week two, and nobody can tell you which step regressed. Shipping evaluation and release gating as a use case suggests the team has been there.

Self-healing browser automation

Playwright automation with auto-heal is one of the named use cases. Browser automation breaking on a selector change is the oldest problem in the category, and an auto-healing layer is a sensible thing to build into a workflow engine rather than leaving to each user.

How Much Does Heym Cost?

Heym is free to use. The site states it plainly: no licensing fees, no usage limits, self-host with Docker Compose, a prebuilt container image, or Kubernetes. For the self-hosted product, that is the complete pricing story and there is nothing else to pay.

What is not published is anything you would pay for. A Cloud offering is listed in the navigation and marked “Soon”. An Enterprise tier exists with its own page, and we checked it: there are no currency figures on it at all, and the calls to action are “Request a demo” and “Explore Enterprise”. So the commercial side of Heym is entirely unpriced today.

Compared to n8n, which offers a free self-hosted Community Edition and then paid cloud tiers metered by full workflow executions rather than by step or by user, Heym’s free tier is more permissive but its paid path is invisible. Dify follows a similar shape, with a free self-hostable Community Edition alongside paid cloud plans. We are deliberately not quoting figures for either, because neither published a number we could confirm directly on their own pages when we checked.

Is Heym Actually Open Source?

No, and this is the thing to understand before you build on it.

Heym is licensed under Commons Clause plus MIT. The site describes this accurately in its footer and licence section: free to use, modify and distribute, but not to resell as a paid service. Commons Clause is not an OSI-approved open source licence, and adding it to MIT removes one of the core freedoms that makes MIT what it is. GitHub’s own API returns a licence classification of NOASSERTION for the repository, which is what happens when a licence file is not recognisable as a standard one.

Heym gets the headline right. The title tag, the hero and the footer all say “Source Available”, which is the correct term. But the body copy slips. Elsewhere on the same page it refers to the “open-source repository” and invites you to “contribute back to the open-source repository”, and describes the platform as built in the open. That inconsistency matters, because the practical restriction is real: if you are an agency or an MSP planning to run Heym on behalf of clients as a paid service, the licence is the first thing your lawyer will stop on.

To be fair, n8n has the same shape with its Sustainable Use License, and this model is now common in the workflow automation category. Heym is not doing anything unusual. It just needs to say “source available” everywhere rather than in most places.

How Mature Is This to Build On?

The repository was created on 28 March 2026 and was last pushed on 9 August 2026, the day before the launch we saw. That is four and a half months old with active development, around 902 stars and 76 forks.

The integration list is long for that age: OpenAI, Anthropic, Ollama, Cerebras, Google Gemini and OpenRouter on the model side, then Slack, Discord, Gmail, Outlook, GitHub, Jira, Linear, Telegram, Notion, Qdrant, ClickHouse, Redis, Supabase, PostgreSQL, Amazon S3, RabbitMQ, WebSocket, Docker, Grist, Google Sheets, Google Drive, BigQuery, Playwright and Sentry. Breadth like that in four months usually means the connector layer is thin, so validate the two or three connectors you actually depend on rather than trusting the logo wall.

One oddity: the repository shows zero open issues. On a project with 902 stars and this much surface area, that is either exemplary triage or a sign that issue tracking is not where the real conversation happens, and the site does point users at Discord for discussion. Neither reading is damning, but if you are evaluating maintenance health, look at the Discord and the commit history rather than the issue count.

The other risk is ordinary and worth stating: this is a two-person team. Self-hosting substantially reduces your exposure, because the code runs on your infrastructure and keeps running whatever happens to the company. That is the strongest practical argument for Heym over a hosted competitor.

Who is Heym Best For?

Use Heym if you are a platform, AI or automation team with a self-hosting requirement, whether that comes from data residency, a regulated industry, or a policy against sending customer data to a hosted automation vendor. If you want agents with approval gates, retrieval and tracing, and you would otherwise assemble that from four separate tools, Heym gives you one deployment with Docker Compose. The price being zero with no usage caps makes it very easy to trial.

Skip Heym if you want a managed service today, because Cloud is not shipping and there is no price for it. Skip it if you plan to resell workflow automation to clients as a paid service, since Commons Clause forbids exactly that. And skip it if your team has no appetite for operating infrastructure, because self-hosting a stack with Postgres, a vector store, a queue and a browser automation layer is real work that a hosted competitor does for you.

Best Heym Alternatives

n8n is the market leader and the honest comparison, which Heym implicitly concedes by publishing a blog post on n8n alternatives. n8n has a free self-hostable Community Edition, a vastly larger connector library, a bigger community and paid cloud tiers billed per completed workflow execution rather than per step or seat. Its Sustainable Use License carries similar resale restrictions. Pick n8n unless you specifically need Heym’s agent-native nodes and built-in evals.

Dify is closer to Heym in ambition, focused on building and operating LLM applications with RAG, agents and observability, and it also offers a free self-hostable Community Edition alongside paid cloud plans. Dify is more mature and better documented. Heym has the stronger visual workflow canvas and more conventional automation integrations.

Langflow or Flowise are the lighter open-source options for wiring LLM chains visually, and both carry properly permissive licences. They are less of a production automation platform and more of a prototyping canvas, with weaker human-in-the-loop and operational tooling. Choose one of these if licence purity matters more to you than runtime features.

Final Verdict: Is Heym Worth It?

Heym is a good piece of engineering with a genuinely permissive free tier and an unusually complete feature set for something four months old. Agent nodes with Python tools, RAG over Qdrant or pgvector, MCP in both directions, approval checkpoints, guardrails, parallel execution and built-in evals is a serious list, and having it all in one self-hosted deployment is worth real money to a team that would otherwise integrate four vendors. If you have a self-hosting mandate, download it this afternoon. It costs nothing and the ceiling is high.

The 7.5 reflects two things a buyer should weigh. Commercially, Heym is opaque: Cloud is “Soon”, Enterprise has no published figures, and the licence forbids reselling it as a service, so if your plan involves paying Heym for anything you cannot find out what that costs. And the marketing calls the project open source in body copy while the licence is Commons Clause plus MIT, which is source available and materially different. Neither problem affects the self-hosted experience, which is free and unrestricted for internal use. Just go in knowing which product you are actually getting, and read the licence before you build a business on top of it.