Summary

Fabro provides expert engineers with a controlled "dark software factory" for AI coding agents. It addresses agent unpredictability by enabling users to define development processes as deterministic workflow graphs, allowing intervention only where critical, thus mitigating risks, fostering collaboration, and optimizing resource usage.

How It Works

Fabro structures AI agent execution via deterministic workflow graphs (DOT-like syntax), supporting branching, loops, and parallelism. Multi-model routing, using CSS-like stylesheets, directs tasks to optimal LLM providers based on cost/capability, including fallbacks. Agents run within isolated cloud sandboxes (Daytona) for security/scalability, offering SSH access. Git checkpointing ensures each stage is version-controlled for traceability.

Quick Start & Requirements

Fabro is a single, compiled Rust binary with zero runtime dependencies (no Python, Node.js, Docker). Install via curl -fsSL https://fabro.sh/install.sh | bash, then fabro install (one-time) and fabro init (per-project). No specific hardware/OS prerequisites mentioned.

Highlighted Details

• Deterministic workflow graphs (DOT syntax) with branching, loops, parallelism, and human gates.
• Multi-model routing via CSS-like stylesheets for LLM selection and cost optimization.
• Isolated cloud sandboxes (Daytona) for secure agent execution.
• Git checkpointing at each stage for traceability.
• Automatic retrospectives detailing cost, duration, and code changes.
• Single binary distribution, requiring no external runtimes.

Maintenance & Community

Fabro uses an issue-based contribution model, discouraging direct pull requests. Users submit issues; maintainers implement changes via AI agents, crediting reporters. Email bryan@qlty.sh for questions. Community interaction via GitHub Issues/Discussions.

Licensing & Compatibility

Licensed under the permissive MIT License, allowing broad usage, including commercial applications and integration within closed-source projects.

Limitations & Caveats

The issue-based contribution model may slow external feature integration. Reliance on external LLM providers introduces dependencies on their availability, performance, and pricing.