Features
Outcome-led, shipped capabilities only: run many repos, preserve state, verify the work, bound cost, and review PRs from anywhere.
Portfolio control
Built for portfolio operators juggling several products, not for single-repo apps. Give every repo a project, run agents in parallel, and watch the fleet from a single overview — all in one browser, no machine to keep awake.
Durable state
Each agent runs in its own isolated sandbox. Your workspace is encrypted and continuously backed up so you never lose work — an idle pause, a restart, or a from-scratch project all survive — and it's never resold, never used for training, and deleted when you destroy the agent.
Evidence
The agent runs your test suite and opens the running app in a real browser to catch console errors and failed requests. A task isn't finished on the agent's say-so — it's finished when the checks back it up, and a failing test sends it back to fix the cause.
Policy
Three things the agent physically cannot do: push to your default branch (rejected at three independent layers), outspend the per-task cap you set (the task is killed mid-stream), or merge without you. Your API keys never touch the sandbox — every model call is brokered server-side. Outbound is fenced at the boundary, with cloud-metadata and internal hosts blocked (verified, not asserted), and logs are scrubbed of keys and secrets before they're ever stored.
Async control
PR notifications land where you already are — Slack, GitHub mobile, email. Your flow doesn't break because you're walking. Approve and merge from the same client you check Twitter on.
Autonomous fleets
Type /loop and hand the agent a standing objective instead of a one-off task. It re-spawns work toward that north star — every few hours on a schedule, or self-paced until the goal is provably met — opening a PR each iteration for you to review. A workflow can be kicked off by you or stood up automatically from a trigger. Every loop is bounded by an iteration count and a total-budget cap on top of the per-task cost cap, so an unattended fleet can never run away. Watch it live, pause or stop it any time.
Models
Every plan includes $100/mo of Kimi K2.6, free. Already pay for Anthropic, OpenAI, Google, OpenRouter, or a custom OpenAI-compatible endpoint? On any paid plan, bring your own key (BYOK) and pay us only the platform fee. Or use our bundled frontier inference, billed at provider cost — usage-based, never marked up. Switch between them per agent.
The shipped surface
It plans, codes, and tests itself
Give it a goal; it breaks the work down and drives it to a PR.
Review and QA, built in
Every plan gets an engineering review, and every agent ships with review passes you can run.
Async-first
Hand off the task and close the laptop.
It learns each repo
The next task on a repo picks up from what already worked, not a blank slate.
See it running, ship it live
Preview the running app in your browser, then publish or deploy it.
Take the wheel when you want to
A real editor, terminal, and source control — edit, commit, and open the PR yourself.
GitHub-native PRs
Branch protection, code review, and CI as designed.
Bring your own guardrails
Block, gate, or get notified at the points the agent can't route around.
See it in the IDE
Self-verifying loop
GitHub-native review
Multi-repo fleet
Get started
Connect a GitHub repo, hand off a task in plain English, and get a reviewed pull request.
free to start · usage billed at cost · cancel anytime