Autonomous AI agents that plan, build, and test your software.
Describe what you want. Auto Code creates the spec, writes the code, runs QA, and hands you a clean branch to review.
Auto Code is a multi-agent framework that turns a plain-language task description into working, tested code. You describe what you want, and a pipeline of specialized AI agents creates a specification, plans the implementation, writes the code, and validates it through automated QA -- all in isolated git worktrees so your main branch is never at risk. A built-in memory system means agents learn from previous sessions and get smarter over time.
|
Planner, Coder, QA Reviewer, and QA Fixer agents work in sequence -- each with a focused role and clear handoff. Every build runs in its own git worktree. Your main branch stays clean until you explicitly merge. Graphiti-powered knowledge graph stores patterns, gotchas, and discoveries so agents improve across builds. A dedicated QA loop catches issues before you ever look at the code, with optional E2E testing via Electron. |
Run up to 12 agent terminals simultaneously. The Coder agent can spawn subagents for parallel subtask work. Import issues, create PRs, and sync progress with your existing project management tools. Works with Claude, OpenAI, Google Gemini, Azure OpenAI, Ollama, and more -- not locked to a single model. Native desktop apps for Windows, macOS, and Linux. Cloud-hosted option also available. |
Kanban Board -- visual task management from planning through completion
Agent Terminals -- multiple AI-powered terminals with one-click task context
Roadmap -- AI-assisted feature planning with competitor analysis
Looking for something specific?
- 📖 Documentation Portal - Comprehensive guide to all documentation
- 📖 Search Index - Comprehensive searchable index with keywords
- 🔍 Search Guide - Learn effective search strategies
- ✅ Verification Checklist - Documentation quality verification
Quick links:
- Quick Start Guide - Get started in 15 minutes
- CLI Usage Guide - Terminal-only usage
- Troubleshooting Guide - Having issues?
Get started in under 15 minutes with our comprehensive Quick Start Guide.
TL;DR: Download → Connect Claude → Open Project → Create Task → Watch agents build → Review & merge
Download the latest release for your platform.
You describe a task
|
v
+--------------+ +-----------+ +--------+ +-------------+ +-----------+
| Spec Creation | --> | Planner | --> | Coder | --> | QA Reviewer | --> | QA Fixer |
+--------------+ +-----------+ +--------+ +-------------+ +-----------+
|
v
You review & merge
Spec Creation analyzes your request and produces a structured specification. The Planner breaks it into subtasks. The Coder implements each subtask (spawning subagents for parallel work when needed). The QA Reviewer validates against acceptance criteria, and the QA Fixer resolves any issues in a loop. You get a clean branch ready to merge.
Desktop (recommended for individual developers) -- Download and run the native app. All processing happens locally.
Cloud-hosted (not ready) – Deploy to your infrastructure for centralized, multi-user access with OAuth, usage tracking, and Kubernetes support. See the Cloud Setup Guide.
For headless operation, CI/CD integration, or terminal workflows:
cd apps/backend
python spec_runner.py --interactive # Create a spec interactively
python spec_runner.py --task "Add auth" # Create spec from description
python run.py --spec 001 # Run autonomous build
python run.py --spec 001 --review # Review changes
python run.py --spec 001 --merge # Merge into your branchSee CLI Usage Guide for full documentation.
Auto Code uses a three-layer security model:
- OS Sandbox -- bash commands run in isolation
- Filesystem restrictions -- operations limited to the project directory
- Dynamic command allowlist -- only approved commands based on detected project stack
All releases include SHA256 checksums. macOS builds are code-signed.
- Discord -- chat, get help, share what you're building
- Issues -- report bugs or request features
- Discussions -- ask questions and share ideas
Contributions are welcome. See CONTRIBUTING.md for setup instructions, code style, testing, and PR guidelines.
Auto Code was originally forked from AndyMik90/Auto-Code. Thank you to the original author for laying the foundation.
AGPL-3.0 -- Auto Code is free to use. If you modify and distribute it, or run it as a service, your changes must also be open source under AGPL-3.0.
