Pilot is the infrastructure for autonomous AI development — async execution, enforced pipelines, a fleet of agents running on a server while you focus on something else entirely.
Every developer using AI today has gone through some version of this progression. Each stage unlocks new leverage — and reveals the next ceiling.
Pilot is built on five infrastructure primitives — the ones that turn AI coding from a productivity tool into a development platform.
Queue a spec, close your laptop, and come back to results. When agents run on a server — not your machine — they report back when they're done. Your schedule and the agents' schedule are fully decoupled.
Multiple specs execute in parallel. One task or ten — the queue handles it. Each agent works in an isolated environment, so they never interfere with each other. Submit from your phone, check results over coffee.
Every task follows the same path: spec → plan → code → review → verify. The pipeline is fully configurable — define the steps that fit your workflow. The system enforces them, not the LLM's memory. Proof is collected at each step, every time, regardless of what the agent decides. Each agent runs in a sandboxed container, isolated from your environment and from each other.
Runs on infrastructure you control — a server, or your own laptop while you're getting started. Remote-accessible from anywhere. No terminal babysitting, no dropped sessions. The work continues whether your lid is open or closed.
Pilot is model-agnostic by design. Use a powerful model for complex reasoning, a faster one for planning, a local model for anything that can't leave the building. Mix and match across the same pipeline — each task can use a different provider if that's what it calls for. Swap models anytime without changing your workflows. Bring Claude, GPT-4, Gemini, Mistral, or run fully local via Ollama. The orchestration layer doesn't care — it just runs the right tool for the job.
Self-host the full orchestration stack on your own infrastructure. MIT licensed, no lock-in, full control.
We host the infrastructure. Your whole team gets orchestration without spending a sprint on DevOps.
The Framework ships when it can build itself. The hosted Team Edition follows. Tell us which you're after — it shapes what we prioritise.
No marketing sequences. Just a heads-up when it ships.
Built in public by Andrew Crookston — engineering leader writing about AI coding at andrewcrookston.com.