Patchwork OS

You don't have an automation problem. You have a decision problem.

Every AI-agent horror story ends the same way: an action nobody stopped to question. Patchwork OS is the layer between the agent's impulse and the action — a local-first runtime where your AI can automate real work across your editor, GitHub, Slack, Gmail, and 45+ services, while anything consequential stops and asks you first.

Three ideas, one runtime:

• Ask before acting. Every action is classified by blast radius — can it be undone, and how much breaks if it's wrong? Reversible things flow freely. Risky things wait in an approval queue with the evidence attached: what exactly will run, why it fired, and what happens if it's wrong.

• Trust is earned, never assumed. Workers — named agents with jobs — start supervised and earn independence from their track record, per action type. Good at filing issues ≠ allowed to push code. One bad high-stakes action outweighs a hundred trivial successes. You set the ceiling; the math never raises it for them.

• Every decision leaves a receipt. What was done, why it was allowed, and how it turned out — durable, replayable, explainable via patchwork judgments, the dashboard's traces page, and patchwork gate explain. When you approve something, you find out later whether you were right.

All of it runs on your machine: your model (Claude, GPT, Gemini, Grok, or local Ollama), your credentials, your logs. Nothing phones home unless you opt in to anonymous analytics.

---

90-second start (no editor required)

npx patchwork-os@beta init   # scaffolds ~/.patchwork, prints your dashboard login
patchwork start              # bridge + dashboard

Open http://localhost:3200. From the browser: run and schedule YAML recipes, connect services (Gmail, Calendar, Slack, GitHub…), review what your agents drafted, and approve — or refuse — anything that wants to leave the machine.

Prereqs: Node 22+. macOS, Linux, and native Windows (no WSL).

The hero workflow — Morning Brief:

patchwork connections connect gmail
patchwork connections connect google-calendar
patchwork recipe run morning-brief

Every morning: a digest of your email, calendar, and overnight agent activity lands in your inbox as Markdown, with any drafted replies waiting for your approval — never auto-sent. No connectors yet? --local runs it against Ollama with your clipboard and recent files.

How it works

Recipes are plain YAML: a trigger (cron, file save, git commit, test run, or any webhook — iPhone Shortcut, Stream Deck, Home Assistant) plus steps. Share them like dotfiles, install them from the marketplace, or let the dashboard generate one from a sentence.

Workers are recipes with an identity and a track record. A worker that triages failing CI starts by only proposing ("this looks like a real break — file an issue?"). Confirm its filings were real and it earns a longer leash — for that job only. It can be demoted in one bad day. You can cap any worker permanently with one line of YAML.

The queue is where impulse meets judgment. Requests arrive sorted by blast radius with evidence inline. patchwork panic stops all automation instantly.

trigger → recipe/worker → [reversible? → run]
                           [risky?      → approval queue → your yes → receipt]

Also in the box: the Claude IDE Bridge

The foundation layer is a standalone MCP bridg