The open-source AI software engineer you can self-host and audit.
For teams that can't send their code to the cloud and won't merge what they can't review. Your infra, your model bill, your audit trail.
Self-hosted on your own Kubernetes via the Helm chart (or docker-compose on a laptop). Your source never has to leave your network — the answer for teams that legally can't ship code to a third-party cloud.
Every run starts from a written spec with acceptance criteria. You approve the plan before code is written and the diff before it merges; a QA agent checks the work. No unsupervised black box, no vibe-code to clean up.
Every action is journaled in a hash-chained audit log; every spec, plan, and QA report lives on disk and in version control. The provenance trail your security team and your auditor actually ask for — SOC2 / ISO evidence in the enterprise build.
Route each phase to the right model — Claude, OpenAI, Gemini, Codex, or a local Ollama / OpenAI-compatible endpoint. Provider abstraction means you own your model bill and never depend on a single vendor.
Each task runs in its own git worktree — nothing touches your tree until you merge. Watch the agent work in real time via the Live Console (read-only by default, one-click Attach), with OpenTelemetry tracing across every boundary.
27 MCP tools across stdio + HTTP+SSE transports. Drive AIFactory from Claude Code, Cursor, Continue.dev, or any MCP-aware client — and hand a task off to an autonomous background run with the /handover skill.
The platform moves fast. Here's what landed lately.
AIFactory now builds from a signed Task Contract (the planned acceptance criteria from PFactory) and hands the built branch + deployed URL to TFactory, which verifies it with visible evidence. Independent subtasks build in parallel with per-worker token, cost, and progress tracking.
See the contract →27 task-control tools across stdio + remote HTTP+SSE transports. Drive AIFactory from Claude Code, Cursor, or any MCP-aware client.
Read the design →Type /handover in Claude Code. AIFactory summarises the conversation, creates a spec, and runs the build while you do something else. Comes back as a draft PR.
See the workflow →A catalog of well-maintained MCP servers auto-enables per project when infra markers + credentials line up. Read-only by default. CVE-aware version pins.
Setup guide →Wires Claude Code's native --remote-control flag into the agent spawn. Open claude.ai/code on any device and pick up where you left off.
Operator guide →Self-hosted, MIT/GPL-3.0 dual-licensed. Bring your own Claude OAuth seat (or run it against Ollama). Zero lock-in.