Cloudflare OS HomeA local, agent-first workspace

Cloudflare OS + LiteLLM + Docker Compose + Tailscale

Get started

日本語

Research Lab

View on GitHub

🧩

Agent-first by design

Create Gadgets, write files, execute code, and review pending changes.

🔌

Bring your own models

Route an OpenAI-compatible model through the project-local LiteLLM service.

🛡️

Private by default

Keep the lab local or expose it only to your tailnet through Tailscale Serve.

📸

Research companion

Follow the separate public lab for experiment records, screenshots, and evidence.

Cloudflare OS Home

An unofficial, reproducible local lab for exploring Cloudflare OS as a workspace where agents can build and run small applications.

This repository keeps the runtime practical: the upstream source is included, LiteLLM is wired inside the Compose network, and the most important agent behavior is backed by a real Gadget smoke test.

Start with a goal

Get started — bring the stack up from a clean checkout.
Usage — register a model, chat, and ask for agentic work.
Architecture — understand which container owns what.
Evidence — understand the split between runtime QA and the research lab.
Research Lab — read experiment records and inspect screenshots.
Troubleshooting — recover from the common local failures.

The short version

text

Browser → Cloudflare OS → project-local LiteLLM → configured model provider

Cloudflare OS supplies the workspace and agent loop. LiteLLM supplies the model routing. This is a local integration, not an official Cloudflare-hosted service.