An agent fleet runs a verified build → verify → fix loop on your repos and ships merge-ready PRs on your own provider keys. Unlike feature-autopilot tools that hand you unverified diffs, automaintainer only ships work that passed its gate.
Write a focus in plain English — or pick a playbook. The agent team scopes it, does it, verifies it passes, and opens a reviewable PR — every run, on your schedule.
Your 2–3 senior devs are constantly context-switching between features, infra fires, and tech debt. Tests are a prayer. Docs are lies. Nobody has sprint capacity to fix it.
Junior devs need oversight that costs more time than it saves. Senior contractors are expensive and still need context. You need output, not onboarding.
You have open PRs from last sprint that nobody’s looked at. Every hour unreviewed is an hour of merge-conflict debt accumulating silently.
AM was built for exactly this. Here’s what it can take off your plate right now — and where it’s headed next.
These are the tasks AM is already doing reliably in production — bounded, verifiable, repeatable work that your team keeps deferring.
The feature/architecture quadrant requires context depth AM doesn’t have yet. Here’s what’s on the roadmap to get there:
Specialised agents — architect, dev, and QA — coordinate over a peer message bus, then every change runs the Verification Gate (build → test → fix → re-verify). Only work that passed its own gate becomes a PR.
Set a cron-style window. Or trigger a run instantly from the CLI or REST API.
A separate review team reads diffs and checks CI on every PR (Pro). On Max it merges approved PRs for you — fully hands-off.
Open issues are injected into every run. Agents fix them and close them automatically with Fixes #N commits.
JSON by default. Semantic exit codes. Structured errors. Designed for AI agents to drive, not just humans.
In hands-off mode (Max): from repo registration to merged PR with no human in the loop. On Free & Pro you review before merge.
Each run gets a dedicated branch and isolated worktree. Two concurrent runs on the same repo can never overwrite each other — even if the previous run never finished.
“You shouldn’t be prompting coding agents anymore. You should be designing the loops that prompt them.” The catch: today that means wiring five primitives together yourself and babysitting the glue. Automaintainer ships them assembled — one hosted, verified loop across a fleet of workers and models.
Scheduled runs + explore-mode discovery find the work and triage it on a cadence. Nobody is pressing go.
Every run gets its own isolated git worktree, so agents working in parallel never step on each other’s files.
Playbooks + per-repo conventions capture project knowledge, so the loop compounds instead of re-guessing from zero each run.
Opens PRs, closes issues, runs your verify command, and (via MCP-capable harnesses) reaches the rest of your stack.
Architect / dev / QA teams — and a separate verifier from the maker, so no agent grades its own homework.
Per-repo memory + a run ledger live on disk, outside any conversation — so tomorrow’s run picks up where today’s stopped.
Build the loop. Stay the engineer.
A loop running unattended is a loop making mistakes unattended — unless “done” means something. AM’s verification gate is a different agent from the maker, and a red gate never ships. You decide review vs. auto-merge. No slop, no cognitive surrender.
Every model is trained differently — so put the right one on each seat. A frontier model to architect, a strong coder to implement, a fast skeptic to verify. Mix tiers to balance quality against token cost, and swap any of them anytime. Your keys, no markup.
Just shipped — Mythos’ Fable 5: assign it to any role the day you have a key.
You pay for orchestration — not inference. automaintainer is bring-your-own-keys: your worker runs the AI harness with your own provider key, so you pay tokens directly with no markup. That's why it's €19, not $200+.
Add your first repo in under 60 seconds. Maintenance wedge today — autonomous improvement on your schedule.
Get started free — no card requiredNeed a human in the loop? The founder is available for architecture reviews, custom integrations, and team ramp-up. Check pricing at Intrane →