---
title: defract — run your whole process, not just your agents
description: defract runs your software process — story, design, architecture, review, release — on parallel Claude Code agents in their own git worktrees. Own your product end to end, without the slop. Open beta for macOS.
doc_version: "1.0"
last_updated: 2026-06-29
canonical: https://defract.dev/
---

# defract — run your whole process, not just your agents

Markdown mirror of [https://defract.dev/](https://defract.dev/).

> defract runs your software process — story, design, architecture, review, release — on parallel Claude Code agents in their own git worktrees. Own your product end to end, without the slop. Open beta for macOS.

## What it is
defract runs your whole software process — scoping, design, architecture, implementation, review and release — in one integrated UI, on parallel Claude Code agents, with a memory system that learns your codebase. You own the product end to end; agents do the work.

**Competitors run your agents; defract runs your process.** One agent is the engine; defract is the chassis.

## The pipeline
A story moves through a gated lifecycle:
1. **Story / scope** — you approve what gets built.
2. **Design** — a built-in visual design stage (uncommon in this category).
3. **Architecture** — you approve the approach.
4. **Implementation** — agents build, each in its own git worktree.
5. **Review** — agents review code; merges only happen after sign-off.
6. **Release** — ship.

You stay above the diff: humans approve scope, design and architecture, then see the working feature. Code only merges after agents have reviewed it.

## Why it's different
- **Opinionated lifecycle**, not just a parallel runner.
- **Built-in visual design stage** — nearly unique in the market.
- **Local-first economics** — PTY transcripts never leave the device; you bring your own Claude and pay Anthropic directly.
- **Worktree isolation** — every story runs in its own git worktree, so parallel agents and people never collide on main.
- **Memory that compounds** — defract learns your codebase across builds.

## FAQ
**What is defract, exactly?** It runs your whole software process — scoping, design, architecture, implementation, review and release — in one UI, on parallel Claude Code agents, with memory that learns your codebase.

**How is it different from Claude Code in a terminal?** Claude Code is a great CLI. defract is the workflow on top — multi-stage, multi-agent, parallel, with embedded memory and git-worktree orchestration.

**Do humans review code, or agents?** Agents review code. Humans approve scope, design and architecture, then see the working feature. Code only merges after agents sign off.

**What stops agents stepping on each other?** Every story runs in its own git worktree — parallel work is isolated by default.

**What's in the open beta vs v1?** Open beta: Claude Code on macOS, downloadable now. v1: Claude Code + Codex + Gemini + opencode, on macOS + Windows.

**What if I want to control every step?** Switch the story to human-in-the-loop — agents pause at every stage and hand you the wheel.

## Get it
Open beta on macOS (Apple Silicon + Intel): [download](https://defract.dev/download). Windows in progress.

## Sitemap
- [Home](https://defract.dev/) · [markdown mirror](https://defract.dev/index.md)
- [Download (macOS)](https://defract.dev/download)
- [Glossary](https://defract.dev/glossary) · [markdown](https://defract.dev/glossary.md)
- [Comparisons](https://defract.dev/vs/conductor): vs Conductor, Claude Code, Agent Teams, Cursor, Kiro, Windsurf
- [Blog](https://defract.dev/blog/)
- [Full sitemap](https://defract.dev/sitemap.md)