Install self-hosted apps, local models, and AI agents in one click, and run them side by side on a Mac, PC, or Linux box.
Your apps, your models, and your data stay on your hardware.
▶ A thirty-second tour. Select it for the version with sound.
What is Companion Hub?
Companion Hub is the local app runtime of the Companion Intelligence platform. It installs and supervises apps as Docker Compose deployments, and gives you a web dashboard, a desktop app, and a cihub CLI to run them.
You can put almost anything in a container and run it here. A whole guest OS. A wealth dashboard you wrote in two prompts. A git repo you just cloned. Hub installs it, watches it, and puts it next to everything else you already run. That is the multi-app world, in one interface, on hardware you can see.
Hub is the data cube. Your data nugget server. The core is free to inspect and run for personal and nonprofit use.
Screenshots
Features
Hub is not a single app. It is the layer that holds the rest together.
Install
Prerequisites: Docker is required. Install Ollama too if you want local models. Check the system requirements for hardware.
Or install from a package manager:
# macOS (Homebrew)
brew tap companionintelligence/homebrew-tap
brew trust companionintelligence/homebrew-tap # first time only, on recent Homebrew
brew install --cask companion-hub
# Windows x64 (Scoop)
scoop bucket add companionintelligence https://github.com/companionintelligence/scoop-bucket
scoop install companion-hub
Open Companion Hub. The setup wizard runs on first launch, starts the Docker stack, and opens the dashboard at http://localhost:5002.
Install the same Linux package from GitHub Releases and start the stack without a window:
curl -fsSL https://get.docker.com | sh # if Docker is not installed
sudo dpkg -i Companion.Hub_*_amd64.deb # or the .rpm / arm64 build
companion-hub --detached
Then open http://<server-ip>:5002. On a public server, prefer Tailscale or the Cloudflare Gateway over open ports.
After install, create your admin account, finish the onboarding wizard, and pair the Hub with your CI Account. The full walkthrough is Installation, or Your first hour for one end-to-end page.
Command line
The desktop app bundles the cihub CLI and copies it into your user bin directory on first launch, so the Hub is scriptable from any terminal.
cihub wizard # launch, configure, and register in one guided flow
cihub up --detached # start the Hub stack
cihub status # containers, tunnel, VPN, and models
cihub app add my-app nginx:alpine --port 8080:80 # run any image next to your apps
cihub api-key create --scope inference # a key for your editor or SDK
See the CLI reference, or docs/CLI.md for every command.
Connect your tools
Hub speaks the OpenAI API and the Ollama native API. Create an inference-scoped key in Settings → Security (or cihub api-key create --scope inference) and point any OpenAI-compatible client at:
http://<hub-host>:5002/api/inference/v1 model: "auto"
The key reaches the inference routes and nothing else. If you already run vLLM or oMLX, set VLLM_URL or OMLX_URL (or the URL in Settings → AI) and the Hub serves from it rather than asking you to switch engines. See docs/editor-inference.md.
Agents connect over MCP. Claude Code, OpenClaw, Hermes, and any client that speaks Streamable HTTP can manage the Hub and read Companion Memory. See Connect an agent.
How it fits together
flowchart LR
clients["Editors and agents<br/>Cursor · Claude Code · Zed"]
subgraph box["Your machine"]
direction TB
hub["Companion Hub<br/>dashboard · desktop app · cihub CLI"]
apps["Apps<br/>marketplace · Compose · git repos"]
agents["Agents<br/>OpenClaw · Hermes"]
memory[("Companion Memory")]
models["Local models<br/>Ollama · vLLM · oMLX · Lemonade"]
hub --> apps
hub --> agents
hub --> models
agents <--> memory
end
portal["CI Account<br/>pairing · sign-in · public web"]
clients -- "OpenAI API · MCP" --> hub
portal -. "device key" .- hub
Hub runs on your hardware. The CI Account is the door for sign-in, the public web, and paid capacity. Add-ons are not a lock on this repo. Read the architecture reference for the full picture.
Documentation
User documentation lives at docs.ci.computer.
Engineering notes for this repository live in docs/. Start with the doc map in docs/README.md.
Security and privacy
- Hub and Portal. Open-source Hub is an untrusted client of Companion Portal. Marketplace installs and registry tag lists need a Portal-issued device key, so pair first (
cihub register or the onboarding UI). An unpaired Hub gets 401 on those calls, and the catalog can look slow or empty instead of obviously unauthorized. See docs/security/hub-portal-trust.md.
- Telemetry. Distributed builds report crashes to Sentry by default.
CI_TELEMETRY=off or CI_LOCAL_ONLY=true in the Hub .env stops it, as does the switch in Settings → General. What is collected, and what is not, is in docs/telemetry.md.
Supported desktop installers
| Platform | Install route | Update route |
|---|
| macOS (Apple silicon and Intel) | Homebrew Cask or release DMG | brew upgrade --cask companion-hub or install the new DMG |
| Windows x64 | Scoop bucket or release installer | scoop update companion-hub or install the new release |
| Linux | Download the Linux release from the website | Install the newer release package |
The package-manager repositories track Hub releases. See Homebrew and Scoop for their platform-specific behavior.
Contributing
pnpm install
pnpm run local
pnpm install is enough to build and run the web stack. The desktop app needs GTK and WebKit headers. That failure and its fix are in docs/DEVELOPMENT_SETUP.md.
Community
Discord · YouTube · X · LinkedIn
License
Companion Hub uses the PolyForm Noncommercial License 1.0.0. You can use, modify, and redistribute it for personal and nonprofit purposes. For common questions, see the License FAQ or email [email protected].
Required Notice: Copyright LifeScope INC, DBA Companion Intelligence (https://ci.computer)
Companion Hub is derived from Runtipi, and would not exist without it. See NOTICE.md and CREDITS.md.
You never change things by fighting the existing reality. To change something, build a new model that makes the existing model obsolete.
R. Buckminster Fuller
