# Toban

> AI agent orchestration platform. Assemble AI agents as a team and run your development in sprints.

Toban orchestrates AI coding agents into a team — Manager, Builder, Reviewer, Strategist — and runs sprints covering planning, development, review, and retrospective. You set the direction, agents execute, and you approve or reject from a web dashboard.

## Docs

- [Getting Started](https://app.toban.dev/docs/getting-started): Quick start guide — GitHub login, project setup, CLI connection
- [Setup Guide](https://app.toban.dev/docs/setup-guide): Detailed setup walkthrough
- [Architecture](https://app.toban.dev/docs/architecture): System architecture — Dashboard (Next.js), API (Cloudflare Workers), CLI (Node.js)
- [API Reference](https://app.toban.dev/docs/api-reference): Full REST API documentation
- [Playbook Guide](https://app.toban.dev/docs/playbook-guide): Agent behavior rules (locked/overridable/recommended)
- [ADR Guide](https://app.toban.dev/docs/adr-guide): Architecture Decision Records for agent context
- [Error Reference](https://app.toban.dev/docs/error-reference): API and CLI error codes
- [Security](https://app.toban.dev/docs/security): Security model and threat mitigations
- [Release Notes](https://app.toban.dev/docs/releases): Version history and known issues

## Key Concepts

- **Workspace**: A project with connected GitHub repo, agents, and sprints
- **Sprint**: Time-boxed work cycle (active/review/retrospective/completed)
- **Manager AI**: LLM that proposes tasks, coordinates agents, runs sprint planning
- **Builder Agent**: Executes tasks in git worktrees, commits code. Multiple builders can run in parallel
- **Reviewer Agent**: Independent code review process with test execution and quality scoring (0-100)
- **Strategist Agent**: Sprint planning, improvement proposals, goal evaluation
- **Product Goals**: Set goals with key results, linked to sprints. Agents are goal-aware
- **Playbook**: Rules that govern agent behavior, with tag-based skill injection
- **Failure Database**: Tracks rejected tasks, patterns injected into prompts to prevent recurrence
- **Shared Agent Memory**: Cross-agent knowledge sharing with tag-based injection
- **Defense Report**: Monthly summary of issues caught by Playbook rules

## Available Now

- Sprint management with kanban board and drag-and-drop
- `toban init` — interactive CLI setup wizard with browser-based GitHub auth
- Parallel builder dispatch with git conflict prevention
- Task auto-split for large tasks
- Dependency-aware task assignment
- Reviewer agent with structured review reports
- Diff analysis — tracks what changes turn rejections into approvals
- Product Goals with key results and strategy log
- Sprint analytics, velocity tracking, and quality scores
- Failure Database with pattern analysis
- Auto-review on git push (pre-push hook)
- Public metrics page for Build in Public
- Agent concurrency settings (max builders, max cloud engineers)

## Coming Soon

- Multi-engine support (Codex, Gemini)
- External skills import
- Team/multi-user features

## Quick Start

1. Sign in at https://app.toban.dev with GitHub
2. Create a project and connect a repository
3. Install Claude Code CLI locally
4. Run: `npx toban-cli@latest start --api-url https://api.toban.dev --api-key <KEY>`
5. Or use `toban init` for interactive setup

## API

Base URL: `https://api.toban.dev`
Auth: `Authorization: Bearer tb_<workspace>_sk_<key>`

Full API reference: https://app.toban.dev/docs/api-reference
Complete docs: https://app.toban.dev/llms-full.txt