Guided setup
2.0 detects an existing ChatGPT or Claude subscription, your API keys, or a local model and wires it up for you. No config files to hand-edit.
OpenClaw is the open-source, self-hosted AI assistant trusted by 385,000+ developers. Version 2.0 rebuilds onboarding, the browser app, sessions and permissions from the ground up.
100% open source · MIT licensed · Runs on macOS, Linux, Docker & WSL2
Measured at the OpenClaw 2.0 launch, 31 August 2026
OpenClaw 2.0
The largest release in the project’s history — roughly half of all code ever merged shipped in this one version.
2.0 detects an existing ChatGPT or Claude subscription, your API keys, or a local model and wires it up for you. No config files to hand-edit.
The Control UI was rewritten from scratch. Interactions now land in about 575ms, with sessions, memory and tools in one place.
Conversations moved onto SQLite. History, memory and context survive restarts instead of living in fragile JSON files.
Share an assistant with your team, run several agents side by side, and keep permissions and secrets under your control.
Quick start
One command gets the whole stack on your machine. 2.0 then walks you through connecting a model.
A single command sets up the runtime on macOS, Linux, Docker or WSL2. No manual dependency hunting.
2.0 detects an existing ChatGPT or Claude subscription, your API keys, or a local model such as Ollama — and configures it.
Start your first session in the rebuilt browser app. Sessions persist, so you can close the tab and come back.
$ curl -fsSL https://openclaw.ai/install.sh | bashmacOS · Linux · WSL2 · Docker
Prefer to read first? Installation guide or quick start.
Free tools
Three small utilities we built while running OpenClaw ourselves. No signup.
Answer a few questions and get a compose file wired for your model provider and storage.
Open toolModel what a month of real usage actually costs before you commit to a provider.
Open toolGenerate a permission and secrets policy for an assistant that touches your files.
Open toolFAQ
It is the largest release in the project’s history — roughly half of all code ever merged shipped in this one version, from 933 contributors across 16,000+ pull requests. The installer and first-run flow were rebuilt, the browser app was rewritten (the Control UI now responds in about 575ms), sessions moved onto SQLite, and permissions, secrets, sharing and multi-agent work were all reworked.
Yes. The core is open source. You pay for your own hardware and, if you use one, for a hosted model API. There is no per-seat licence and no usage tax from us.
No. A Mac Mini is a popular always-on host because it is quiet and efficient, but 2.0 runs on macOS, Linux, Docker and WSL2. Pick whatever machine you already have.
Yes. Guided setup in 2.0 detects an existing subscription, API keys you already have, or a local model such as Ollama, and wires it up without hand-editing config files.
It stays on the machine you installed it on. That is the entire point of self-hosting — you choose which model provider, if any, sees a request. Nothing is proxied through us.
There is a dedicated migration guide. In practice it is mostly renaming configuration files and updating the image tag.
Still stuck? Read the docs or see the migration guide.
One command, about five minutes, and your assistant runs where you can see it. 385,000+ developers already made the switch.
Looking for the most secure AI assistant? Openclaw is the answer.