https://github.com/kagan-sh/kagan

The Orchestration Layer for AI Coding Agents
https://github.com/kagan-sh/kagan

ai ai-agents antrophic claude claude-code cli codex gemini kagan kanban mcp openai opencode python textual tui uv

Last synced: 27 days ago
JSON representation

The Orchestration Layer for AI Coding Agents

• Host: GitHub
• URL: https://github.com/kagan-sh/kagan
• Owner: kagan-sh
• License: mit
• Created: 2026-01-25T22:09:11.000Z (4 months ago)
• Default Branch: main
• Last Pushed: 2026-04-30T20:06:30.000Z (about 1 month ago)
• Last Synced: 2026-04-30T20:19:47.815Z (about 1 month ago)
• Topics: ai, ai-agents, antrophic, claude, claude-code, cli, codex, gemini, kagan, kanban, mcp, openai, opencode, python, textual, tui, uv
• Language: Python
• Homepage: http://kagan.sh/
• Size: 12.7 MB
• Stars: 66
• Watchers: 2
• Forks: 5
• Open Issues: 2
• Metadata Files:
  • Readme: README.md
  • Changelog: CHANGELOG.md
  • Contributing: CONTRIBUTING.md
  • License: LICENSE
  • Code of conduct: CODE_OF_CONDUCT.md
  • Codeowners: .github/CODEOWNERS
  • Security: SECURITY.md
  • Agents: AGENTS.md

Awesome Lists containing this project

• awesome-ai-devtools - Kagan - source AI orchestration board with a VS Code extension for planning, running, and reviewing coding agent tasks. (Agent Infrastructure / Multi-Agent Orchestration)
• awesome-tuis - kagan - powered Kanban TUI for autonomous development workflows (Table of Contents)

README

          


  

    

    

    

  





  

  

  

  

  





  

  

  

  





  Docs ·

  Quickstart ·

  The review gate ·

  MCP Setup ·

  CLI Reference ·

  Contributing



---

Kagan is a Kanban TUI for AI coding agents with a structural human review gate. No agent-authored task reaches your main branch without an explicit approval — the state machine enforces it.

The agent runs in an isolated git worktree. When it finishes, the task card moves to REVIEW. You read the diff, check the acceptance criteria, and press approve. Then merge fires. That transition — REVIEW to DONE — cannot be automated away. It is not a setting.

## Install

```bash

uv tool install kagan     # or: uvx kagan

```

Mac / Linux (no uv)

```bash

curl -fsSL https://uvget.me/install.sh | bash -s -- kagan

```

Windows (PowerShell)

```powershell

iwr -useb uvget.me/install.ps1 -OutFile install.ps1; .\install.ps1 kagan

```

## What you get

- Kanban board (BACKLOG → IN_PROGRESS → REVIEW → DONE) enforced by a state machine

- Each task runs in its own git worktree — your working copy stays untouched

- Managed runs (background agent) or interactive attach (you + agent in tmux/editor)

- REVIEW stage requires explicit human approval before merge; no path around it

- MCP server so Claude Code, Codex, or any MCP-capable client can drive the board

- `kagan doctor` preflight checks all required tools before first run

Tested agents: **Claude Code** · **Codex** · **Gemini CLI** · 11 more — see [docs/backends](https://docs.kagan.sh/concepts/architecture-overview/#supported-agents).

Full docs: **[docs.kagan.sh](https://docs.kagan.sh/)**

## Companion surfaces

The TUI (`kagan`) is the primary operator surface. Two companion surfaces exist for specific workflows:

- **Web dashboard** (`kagan web`) — browser-based board, useful for remote access or a second monitor

- **VS Code extension** — sidebar panel and `@kagan` chat participant inside VS Code

Both companions share the same state as the TUI via the same API server. Neither is required.

## License

[MIT](LICENSE)

---