AI devs · Fleet

An AI dev takes the engineering. You keep the ideas.

It scouts the repository, plans the work, writes the code, opens the pull request, reviews it, chases red CI and merges through machine gates, around the clock. The code is written on your machines, on your keys.

Request access See how it works

Audited end to endGitHub-native workflowExisting checks stay in controlCost visible per task

Live fleet

See work moving, not tickets waiting.

Every task in the fleet has an owner, a stage, a policy boundary, and an audit trail attached, across every repo the app is installed on.

Fleet / example3 active

api-core · #482Upgrade zod across servicesweb-app · #519Fix auth redirect loopdocs · #204Refresh the quickstart

Agent 01

Upgrade zod across services

Read repository context
Proposed implementation plan
Running checks

Agent 02

Fix auth redirect loop

Read repository context
Proposed implementation plan
Writing tests

Agent 03

Refresh the quickstart

Read repository context
Proposed implementation plan
Opening pull request

What an AI dev does

01
Scout
Reads the repository, the thread and the history, then writes the plan down, so you can review the shape of the work before any code exists.
02
Triage
Routes the work to a lane, or declines it with a reason you can read.
03
Worker
Writes the code on your own box, on your own keys, and opens the pull request.
04
CI + reviewer
Reviews the diff, answers the thread and chases red CI until the checks are green.
05
Merge lane
Merges only past the machine gates. A human lane stays available, per path or per branch.

Two things are deliberately not on this list. The ask arriving is yours, and the deploy after the merge is your own pipeline: we never ship your software for you.

The output

Pull request #482

fix: refresh expired installation tokens

Ready to merge

+ if (installation.tokenExpiresAt < now) {
+   await refreshInstallationToken(installation)
+ }

6 files changed · +84 −19Policy checks passed

An illustration, not a receipt: the pull request number, the diff and the file count are all made up.

Who does what

Work handed from role to role

Triage judges each filed ask on your key, and drops some with a reason. A plan your agent starts is written by the scout and goes straight on. The worker writes the pull request on your box, the reviewer sends some back for a fix, and the merge gate passes each one or holds it for a human.

in flight
waiting on a human
dropped with a reason
merged, in your pipeline

An illustrative animation, not a live feed: the work and its timing are made up, and nothing here is measured from our fleet.

Built for teams that ship

BYOK: bring your own keys
Connect Anthropic, OpenAI, Gemini, DeepSeek, Kimi, GLM, MiniMax and more, or any OpenAI-compatible endpoint. We never proxy or resell tokens. Your model bill, your call.
Your hardware (BYOC)
The coding and the CI run on boxes you enrol: your VPS, your metal. No cloud compute markup.
Zero quotas
No task caps, no file limits, no per-seat add-ons. One flat price per AI dev. Everything included.
Every action audited
Every tool call, decision, and merge gate is logged with full reasoning. If it isn't in the audit trail, it didn't happen.
Never argues with other bots
CodeRabbit, Copilot, Dependabot and Renovate are recognised by identity and never answered, so two bots never end up arguing on one thread. Dependabot and Renovate PRs are still reviewed, fixed and merged through the merge gate.
Policy, not state machines
Drop a .dz/maintainer/maintainer.yml in your repo. The agent reads it as policy, no hardcoded workflows, no hidden defaults.