Fabro
Fabro is an open source “dark software factory” that gives engineers a deterministic harness around AI coding agents. Instead of babysitting every agent step or reviewing an opaque 50-file diff, you define your process as version-controlled workflow graphs in Graphviz DOT and let Fabro orchestrate AI agents, shell commands, and human decisions into repeatable, long-horizon coding processes.
Key Features
- Workflow-as-code — Define pipelines in Graphviz DOT with branching, loops, parallelism, and human gates — diffable, reviewable, and version-controlled.
- Human-in-the-loop — Approval gates pause for decisions, steer running agents mid-turn, and interview steps capture structured input at the right moments.
- Multi-model routing — CSS-like stylesheets route each node to the right model and provider, with automatic fallback chains when a provider goes down.
- Cloud sandboxes — Run agents in isolated cloud VMs with snapshot-based setup, network controls, and automatic cleanup; SSH in with
fabro ssh. - Git checkpointing — Every stage commits code changes and execution metadata to Git branches, so you can resume, revert, or trace any change.
- Automatic retrospectives — Each run generates a retro with cost, duration, files touched, and an LLM-written narrative to improve workflows over time.
- API server + web UI — REST API with SSE event streaming and a React web UI; queue runs, track progress, and manage at scale.
- Single binary, no runtime — One compiled Rust executable with zero dependencies — no Python, Node, or Docker required to run it.
Why Use Fabro?
AI coding agents are powerful but unpredictable. Fabro replaces the prompt-wait-review loop that idles engineers and demands constant babysitting with deterministic, version-controlled workflows. You intervene only where it matters, route each task to the most cost-effective model, and capture every tool call, agent turn, and shell command in a unified observable event stream — queryable with SQL via DuckDB.
Use Cases
- Plan–approve–implement–simplify pipelines where a human approves each stage before agents continue.
- Running long-horizon coding processes unattended, with agents working across cloud sandboxes in parallel.
- Building an evaluation and verification suite (LLM-as-judge, test suites, human review) that grows confidence over time.
Platform
Linux · macOS · Windows (single Rust binary) · Self-hostable via Docker/cloud
Licence
MIT