Wake up to pull requests to merge, and your next customers already found.

Claude Code is an agent that acts. 1mn wraps that with the rest of the loop — discover, decide, verify, gate, persist, hand off. The hard part isn't the agent acting; it's knowing when not to act, how to check the output mechanically, and how to surface the right thing to you at the right moment. That's what the loop shape provides.

It depends on verifiability and blast radius. PRs go through your normal GitHub review — the loop opens them, you merge. Support replies are drafted and queued — you approve before send. SEO and content are recommend-only by default. Ads can be set to autonomous pausing but never autonomous spend increases. Every gate is explicit and configurable.

Yes. However you built it — Lovable, Bolt, v0, Cursor, Claude Code, Base44 — 1mn works after it. The one requirement is a GitHub repo, and messy, test-free, AI-generated code is fine. It reads your repo and infers the stack (Cloudflare Workers, Vercel, Supabase, Drizzle, Prisma, tRPC, and most common solo-founder setups). If it can't infer something, it flags the assumption before acting — not after.

Three connections, all up front (nothing hidden behind the paywall): 1) connect your GitHub repo via our one-click App so the loops can read your code and open PRs; 2) install the 1mn SDK — a single snippet — so the loops see real errors, feedback, and usage (or let 1mn open a PR that wires it in for you); 3) connect Claude Code with your own API key or Pro/Max subscription. Don't want to deal with the Claude Code wiring? 1mn does that setup for you as its first ticket.

Opinionated means the loops ship with sensible defaults: where the human gates sit, what verification looks like, what the trust ladder is. You can adjust gate thresholds and scope per loop. What you can't do is disable verification entirely or remove the human gate on irreversible actions — that's load-bearing.

It runs in a sandbox, not your production branch. Every change opens a PR — a proposed change you review before it goes live — and the loop won't touch auth, billing, migrations, or anything you've marked off-limits. It runs your test suite first: if a fix looks right but breaks a test, the PR never opens. If it opens one you don't like, you close it — nothing shipped.

Not in draft-only mode (the default). It queues replies for your review and you send. You can graduate specific ticket classes to auto-send after the loop has earned that trust — but it's opt-in, one class at a time.

Every action is logged with its reasoning and evidence. Reversible mistakes get caught by the loop's own verify stage — a fix that fails your tests never becomes a PR. Irreversible ones are gated by design: nothing merges, posts, or spends until you say so. If something slips through, you get a full audit trail — and the gate tightens.