Scalable, observable, governed AI SDLC.

With Facility you set up your own AI Software Development Lifecycle (AI SDLC) platform, open source and self-hosted. Stories go in, reviewed pull requests come out, and people hold on to the decisions that matter.

We built it because the bottleneck moved. Agents can produce more code than a team can read; the scarce work now is deciding what gets built and checking what came back.

A story can start anywhere someone touches your product: a sentence in the feedback form, an exception in production, a request in a chat thread. Facility turns the signals worth pursuing into drafted stories, with the project's context attached and duplicates caught. A person chooses what gets built.

A crew of agents carries each story: one plans against the real codebase, one implements in a disposable provisioned environment and runs the checks, others review, repair routine CI failures, and sweep for security problems. Engineers approve plans, steer agents mid-run, review pull requests, and sign every merge; their job becomes brokering many streams of work asynchronously. One person's attention covers far more than their typing ever could.

Isolated agent sessions lose their context the moment they end; Facility keeps it. Decisions, standards and review feedback become structured, human-approved knowledge that every agent reads, so cost per task falls as the record grows and one engineer's answer keeps serving the whole team.

The guardrails are structural. An agent's sandbox holds no credentials and cannot push to your repository; to ship, it files a delivery receipt and Facility makes the signed commit and opens the pull request. Agents cannot approve, merge, or deploy to privileged environments, and everything they read is untrusted data. The capability is simply not there.

Every run leaves a record: how many agent pull requests got accepted, how many needed a human fixup, what each change cost. The numbers come straight out of the pipeline and decide how much autonomy the agents earn.

Trying it commits you to nothing; agents write back only what a human collaborator would, and disconnecting leaves no trace. We use Facility to ship our own products at The Agile Monkeys and published it early so people can use it and argue with it in the open. Expect rough edges.