Open source · AGPLv3 · Self-host or hosted · No crypto

The fastest way to safely expose your local inference to the internet.

Run one command and your Ollama instance (more backends coming soon) becomes an OpenAI-API-compatible endpoint, reachable from anywhere. No open ports, no reverse proxy, nothing to wire up. Your apps just point at a URL and a key.

Total time computed2.4 days
On your machine
bash
curl -sSf https://cocompute.ai/install.sh | COCOMPUTE_URL=https://cocompute.ai bash -s -- --token YOUR_TOKEN

Grab your token after signup, from the dashboard. Linux + macOS, background service.

From any app
connect
Base URL
https://cocompute.ai/v1
API key
9f2a3b1c••••••••••••••••
Works with anything that speaks the OpenAI API spec
Open WebUIAnythingLLMJanLibreChatLangChainContinuen8nyour code

Set the base URL and a key. That's the whole integration.

Safe by design

Your machine stays closed. The orchestrator is the only thing facing the internet.

No open ports

Your machine dials OUT over an encrypted connection (built on iroh). Nothing inbound to port-forward or scan. Your home firewall stays shut.

Your home IP stays hidden

Apps hit cocompute's stable HTTPS endpoint. Your host sits behind it, anonymous, and can move anytime without anyone noticing.

Revocable keys

256-bit API keys, hashed at rest, scoped to a pool. Rotate or revoke in a click. Single-use, expiring tokens claim each host.

Share it when you want

Pool with friends or your team. Share your endpoint, or tap into theirs, in a few clicks.

Coming soonA public marketplace: earn for the compute you share, or pay to burst onto others' GPUs.

Why we built cocompute

Open models win when they're easy

Open-source models are catching up fast, but self-hosting one and reaching it from your apps is still too much work. Every bit of friction we remove is a reason to pick open over a closed API. That is the whole point of the one-command setup.

Compute should be shared

GPUs sit idle everywhere while people pay cloud premiums for the same cycles. We are building toward an open marketplace: share what you have, use what you need, no gatekeeper taking a cut. Not yet, but it is where this is headed.

Open source. AGPLv3.

cocompute is free software you can self-host, fork, and modify. The code that runs cocompute.ai is the same code in the public repo. cocompute.ai is the hosted version for people who don't want to operate their own orchestrator.

Two ways to run cocompute

Same protocol either way. Let us run it, or self-host for full control.

Self-host

Free

AGPLv3 · run your own orchestrator


  • Full source code, nothing hidden
  • Complete control over hosts and pools
  • Run on your own infra, your own domain
  • No telemetry, no third party
  • Modify and extend, changes stay open

cocompute.ai

Free

The easy way · we run it


  • One command to expose a host
  • OpenAI-API-compatible /v1/ endpoint
  • No open ports, no infra to operate
  • Share with friends or your team
  • Paid public marketplace coming soon

Expose your local model in the next five minutes.

Free signup. No credit card.

Sign up

© 2026 cocompute · GitHub · AGPLv3