Agentagents2y ago

OpenHands review

An open, self-hosted control center for autonomous coding agents — own the infrastructure, bring your own model, and run agents that edit files and ship code.

Maker
All-Hands-AI
Launched
Mar 1, 2024
Pricing
open-source
Visit official site
Firstlook

Our verdict

OpenHands has grown from a single open coding agent into the open answer to a real question: who owns your autonomous-coding setup? It self-hosts, brings your own model, and can even drive other agents (Claude Code, Codex) behind one control center — so your code and keys never leave your infrastructure. The trade is operational: you run it. If you want autonomy without handing a vendor your codebase, it's the most credible open option; if you just want a button, a hosted agent is less work.

First look — our read from the docs and sources below; not yet hands-on tested.

The interesting question about autonomous coding in 2026 isn't "can an agent write the code" — several can. It's "who owns the setup that does it." OpenHands, the All-Hands-AI project sitting at 78.4k GitHub stars, is the clearest open answer: a self-hosted control center for coding agents where your code, your keys, and your infrastructure stay yours.

It started life as a single open coding agent and has grown into something broader: a place to run agents — its own, plus Claude Code, Codex, and Gemini — across your laptop, Docker, a VM, or your cloud, switching between them without losing the thread. The agents do real work: edit files, run commands, drive a browser, and ask for permission at each step. You can wire them into Slack, GitHub, Linear or Notion and trigger them on a schedule or a webhook.

The catch is the obvious one for anything self-hosted: you run it. That's the whole trade. A hosted agent is a button; OpenHands is infrastructure you stand up and maintain. In exchange, nothing about your codebase has to leave your walls, and you're never locked to one vendor's model — it's a harness, so its ceiling is whatever LLM you bring.

Who it's for

Teams who want autonomous coding but can't or won't ship their code to a third party — regulated shops, security-conscious orgs, anyone who wants to own the stack. The "bring your own model" design and the ability to orchestrate other agents under one roof make it a genuine control center rather than just another assistant.

Who should skip it

If you want to open your editor and have an agent working in two minutes with zero ops, a hosted tool (or the managed OpenHands Cloud) is less friction — you just give up the self-host advantage. And if you'd let an autonomous agent edit files and run commands without sandboxing and review, skip all of these until you've set up guardrails; the failure mode of an unsupervised agent is not a typo.

This is a first look from the repo and docs, not our own run, so no score yet — but as the open, own-your-infra answer to autonomous coding, OpenHands is the one to beat.

Specs & key facts

What it isSelf-hosted control center for autonomous coding agents[src]
Popularity78.4k GitHub stars[src]
Runs agentsOpenHands, Claude Code, Codex, Gemini — switchable backends[src]
What agents can doEdit files, run commands, use the browser, ask permission per step[src]
ModelsBring your own LLM (any provider); ACP-compatible[src]
AutomationsSlack / GitHub / Linear / Notion; schedule + webhook triggers[src]
Run itnpm global, Docker, or from source; CLI agent-canvas[src]
LicenseMIT (commercial use allowed)[src]

How to use it

  1. 1Self-host it: install globally (npm i -g @openhands/agent-canvas), run via Docker, or build from source.
  2. 2Point it at your own LLM — any provider — so your code and keys stay on your infrastructure.
  3. 3Open a task or an issue and let the agent plan, edit files, and run commands, approving steps as it goes.
  4. 4Wire automations into Slack, GitHub, Linear or Notion, triggered on a schedule or by webhooks.
  5. 5Prefer not to run infra? Use the managed OpenHands Cloud instead — but then you give up the self-host advantage.

Pricing

Open source (self-host)

Free (MIT)

Run it on your laptop, a server, a VM, or your own cloud. You pay only for the LLM you point it at and the compute you run it on.

OpenHands Cloud / Enterprise

Paid

Managed cloud and enterprise tiers exist as alternatives to self-hosting (pricing on the official site).

Open-weight tooling (MIT) with optional managed Cloud/Enterprise. There's no model cost in the tool itself — you bring your own LLM. Verified 2026-06-26.

Pros & cons

Pros

  • Fully open (MIT) and self-hosted — your code, keys and infra stay yours.
  • Bring your own model, and even run other agents (Claude Code, Codex, Gemini) under one roof.
  • Real autonomy: edits files, runs commands, uses the browser, with per-step permission.
  • Huge, active community (78.4k stars) and built-in automations (Slack/GitHub/Linear/Notion).

Cons

  • You run the infrastructure — setup and upkeep are on you (Docker/VMs/cloud).
  • Autonomous agents that edit files and run commands need guardrails; review every step.
  • Quality depends entirely on the LLM you bring — it's a harness, not a model.
  • If you want a one-click hosted experience, the managed tiers cost money and undo the self-host benefit.

Alternatives

FAQ

Sources

Sources

  1. 1.What it is, 78.4k stars, capabilities, backends, install methods, hosting optionshttps://github.com/All-Hands-AI/OpenHandsVerified 2026-06-26
  2. 2.MIT licensehttps://github.com/All-Hands-AI/OpenHands/blob/main/LICENSEVerified 2026-06-26

More coverage

News & first-looks about this release. Coming soon.
Head-to-head comparisons. Coming soon.