Run AI agents for multiple users, with chat, Git repositories, and access to AI models. Octogee Cloud combines OpenClaw, Mattermost, and Gitea into a platform you can host and manage.
Each user gets a chat account and a place to keep their agent's work. Users message their agents in Mattermost and work with files in Gitea. Operators assign users to shared infrastructure or a dedicated gateway and move them between the two as their needs change.
Try it locally · Run the full platform · Contribute
An agent conversation from the GKE installation walkthrough, using Meta Muse Ember through OpenRouter. Click any screenshot to enlarge it.
- OpenClaw runs the agents. An agent combines an AI model with tools to carry out tasks. A gateway is the service that runs agents and connects them to chat and tools.
- Mattermost provides chat. Users send messages, receive replies, and collaborate in channels.
- Gitea stores work. Each user gets an organization with a workspace repository for files and version history.
- Octogee Cloud manages the fleet. The fleet controller assigns users to gateways, moves them between gateways, and coordinates updates. A model broker connects agents to AI providers and records usage.
Shared gateways serve multiple users through platform-controlled tools.
Isolated gateways give one user shell tools and the OpenClaw Control UI in a
dedicated environment. Operators manage both with the
octogee-fleet command line.
Read the architecture guide,
trust boundaries, and
Admin API automation guide.
| What you want to do | Start here | What you get |
|---|---|---|
| Talk to an agent on your machine | Local foundation | One OpenClaw gateway, Mattermost, and Gitea in Docker |
| Explore user management and shared gateways | Full platform on kind | The fleet controller, platform API, model broker, chat, and Git on a local Kubernetes cluster |
| Install the foundation on Google Cloud or AWS | Cloud installation | One gateway, chat, and Git on GKE or EKS with managed storage |
| Run the full platform in a cloud account | GKE or EKS | Fleet management, user accounts, model access, chat, and Git with managed storage |
| Deploy and operate the full platform in production | Production deployment | Ingress, regional capacity, upgrades, and recovery |
| Improve the code or documentation | Contributing | Local package checks and a guide to your first pull request |
The local foundation needs Docker and a model provider API key for agent replies. The full platform uses kind, which runs Kubernetes in Docker; allow at least 8 CPUs and 12 GB of memory for Docker. See requirements for tools and the installation guide for accounts and model access.
For published builds, start from the latest release and follow the guide included in that checkout. Releasing explains image, chart, and package versions.
| Location | Purpose |
|---|---|
core/, core-testing/, shared/ |
Shared runtime, platform state, and types |
fleet-controller/ |
Fleet controller and administration API |
platform-api/, llm-broker/, gateway-edge/ |
Platform API, model access, and gateway routing |
deploy/, gateway/, desktop/, charts/, terraform/ |
Installers, images, and infrastructure |
gateway-supervisor/, provisioning-service/ |
Gateway lifecycle and account provisioning |
scripts/, tests/ |
Operations, build tooling, and cross-package tests |
templates/ |
Starter files for agents and workspaces |
docs/, AGENTS.md |
Guides, engineering conventions, and agent instructions |
Documentation improvements, bug reports, and code changes are welcome. You can run package tests with Node.js 24 and npm, without a cloud account or model key. Contributing walks through setup, checks, and commit sign-off. Our code of conduct applies in project spaces.
Use the security policy to report vulnerabilities privately. The OpenClaw compatibility guide records the build and patches used by the fleet image.
Build main publishes image manifests, vulnerability reports, and software bills of materials. Verify published images explains how to check their provenance. The source and image security workflows run recurring scans.
MIT.
