SaifCTL — safety harness for autonomous AI agents

Spec-driven AI factory. Use with any agentic CLI. Language-agnostic. Safe by design.

Like GasTown, but agents can't cheat, leak, wreak havoc.

Full feature preview at:

⚠️ Status: Alpha. SaifCTL is under active development. See the Roadmap for what's coming next.

Sponsor this project

Stop Coding. Start Spec'ing.

SaifCTL implements state-of-the-art (early 2026) architecture for Agentic engineering.

SaifCTL guarantee:

• The AI builds exactly what you asked for.
  • The agent is locked in a loop and physically cannot stop until your new TDD tests pass.
• The AI can't break previously-built features.
  • All features built with SaifCTL are protected by tests. AI can't break or change them. Regressions are mechanically impossible.
• The AI breaks nothing on your machine.
  • The agent runs in a zero-trust, sandboxed Docker environment. Your existing codebase is safe.

Read more on Security & Isolation.

The Gauntlet: Merge with Confidence

The AI agent is trapped in a rigorous convergence loop. Every time it writes code, it must survive three stages before opening a PR:

1. The Gate: Your linters, type-checkers, and other static analysis tools.
2. The Reviewer: Adversarial AI that scrutinizes the diff to ensure it matches the spec without taking shortcuts.
3. Holdout Tests: Hidden tests. Agent can't see them. Can't fake a pass.

You only get notified when the code emerges victorious.

Batteries-Included

SaifCTL supports out of the box:

• All major LLM providers + OpenRouter + OpenAI-compatible APIs
• 14 Agentic CLI tools
• 4 Programming languages (Node.js, Python, Go, Rust)
• All major Git providers

Deployment

SaifCTL runs as a CLI that spins up coding agents in ephemeral Docker containers on your machine. Self-hosted and Kubernetes (Helm) deployment support is underway.

Try it out now

SaifCTL is currently in active development. The Docker isolation environment and VSCode extension are dropping in a few weeks.

Star the repository to get notified of the Alpha drop, or Join the Design Partner Waitlist to get early access.

Step-by-step guide

See the Step-by-step usage for a detailed walkthrough of the workflow.

For shorter, task-focused tutorials, see Guides.

VSCode extension

The SaifCTL VSCode extension provides a dedicated sidebar panel to manage your entire AI engineering workflow directly from your editor.

What the extension does:

• Manage Features: Visual tree view of your features. Create new features, or manage existing ones through GUI.
• Design & Run: One-click actions to generate specs (saifctl feat design) and start the coding swarm (saifctl feat run).
• Track Runs: A Kubernetes-style dashboard of all your agent runs. See status (success/failed), view run configs, and instantly resume failed runs or clear old ones.

Requirements

• Node.js 22+
• Python 3.12+
• Docker
• Git
• LLM API key
• Linux or MacOS (Windows is not supported yet)

A fully customizable factory

Every component of SaifCTL is fully modular. You can swap, customize, or disable to fit your team's needs:

• Want to use a different LLMs for coding and designing agents? Easy.
• Want to use your custom Playwright setup for testing? Done.
• Need to enforce strict filesystem rules? It's built in.

Dive into the details of what you can customize in the Features guide.

Reference

• Usage
• Guides — Step-by-step tutorials
• Configuration
• Spec-driven development
• Agents
• Security & Isolation
• Access control with Cedar
• Environments and Infrastructure
• Sandbox profiles
• Test profiles
• Semantic reviewer
• Spec designers
• Codebase indexers
• Source control integrations
• Commands
• Agent environment variables and secrets
• Env vars reference

Development

See our Development guides

Clone with submodules:

git clone --recurse-submodules https://github.com/JuroOravec/safe-ai-factory.git

License

MIT