Private beta
A devbox is a coding agent with its own machine. You hand it a task in Slack, Linear, or GitHub. It comes back with a tested pull request.
Free on your own machine · Cloud hosted in Germany, free during beta
stream_unzip and s3fuse to reduce memory usage based on thread01 / What happens
A blueprint is the recipe a devbox follows for one kind of work. The agent takes the parts that need judgment. A harness runs the parts that must not be skipped, like lint, tests, and review. When a step fails, the work stops there instead of landing in your repo. Devboxes ships with ready-made blueprints, and you can build your own.
02 / The queue
Every task gets its own devbox, so ten run as easily as one. You queue them from the tracker you already use, and what goes in at night comes back as pull requests by morning.
03 / On your machine
The listener installs in a minute and connects the subscription you already pay for: ChatGPT, GitHub Copilot, or any API key. Tasks run in containers on your hardware.
Claude Code and Codex can dispatch tasks and pick up the results later.
Tokens are billed by your provider. Devboxes adds nothing on top.
Signed builds, published checksums
04 / Your code
A devbox works on real repositories with real credentials. Here is exactly how that is handled.
*Cloud devboxes only. Local runs keep your code on your machine; only the session event stream reaches our server, so you can resume from anywhere.
05 / Memory
Every session is stored with what it read, what it decided, and why. When a new task touches the same code, the devbox already knows what happened last time, the way a colleague who was there remembers. Ask why uploads are capped and it points to the incident that made the cap necessary.
Opt-in only. It never leaves your organization.
Planned · the memory starts to compound
06 / Trust
A pull request from a devbox is not a black box. Attached to it is the full session: which sources it read, what it decided, and the recording that proves the change runs. Your team reviews and merges the way it always has.
On-prem is a white-glove deployment, at the same rates.
See pricing →07 / Pricing
A cloud devbox costs money only while it works, and the beta includes 20 free hours a month. Tokens stay on your own provider account, with no markup and no credits. The hourly rates below are the ones beta users lock in.
Estimated compute
$20.00 /mo
Figured at about 1 hour of active compute per task. Model charges billed by the provider.
200 tasks × 1 h × $0.10/hr
Dedicated and on-prem. A white-glove deployment, at the same rates.
Contact us →Today a task comes from the dashboard, the API, or the CLI, and the work lands in GitHub. Slack, Google Chat, Microsoft Teams, and Linear are being built during the beta, with Sentry and Alertmanager as sources.
The subscription or API key you connect, billed by that provider. There is no markup.
No. Sessions are stored for you, encrypted, and used for nothing else. Code intelligence is opt-in and stays inside your organization.
That depends on the blueprint. A human keeps the final call before a merge either way.
Locally it never leaves your machine. In the cloud it runs in Germany, encrypted, on your retention schedule, and is never used for training.
There are two: local runs stay free, and cloud devboxes are free during the beta with 20 hours a month included.
Yes. On-prem is a white-glove deployment: we install it with you, at the same rates.
In stages. Signups join a queue, and confirming your email moves you up.
The button copies a plain description of what Devboxes does today and what is still planned. Paste it into ChatGPT or Claude and ask what you would ask us.
Hand a devbox one task and look at what comes back. Access opens in waves while the beta runs.
inline