Tazuna

Tazuna (手綱) — the reins. Hold the reins on AI-driven software delivery.

Tazuna is an open-source platform for orchestrating AI agents through structured software development workflows. You define Tracks (workflows) composed of Stages, which are composed of Tasks — and AI agents execute them under human-in-the-loop approval, with artifacts, hooks, and Git/Jira/Azure DevOps integration along the way.

The name comes from the Japanese word for reins: you don't replace the powerful agent doing the work, you steer it.

Features

• Track / Stage / Task workflow model with reusable templates
• AI agents powered by Claude (CLI or API mode) with autonomous, collaborative, and manual task runners
• Approvals — sign-off gates between stages
• Artifacts — first-class deliverables linked across tasks
• Hooks — event-driven extensibility (pre/post task, stage, track)
• Tools — call external HTTP endpoints from agents
• Integrations — sync with Jira and Azure DevOps
• Git integration — branch and worktree management
• AI track designer — generate workflows from a natural-language description

Tech stack

• Client: React 19, Vite, TypeScript, Socket.IO client
• Server: Express 5, Knex + PostgreSQL, Socket.IO, Anthropic SDK
• AI: Claude (via CLI subprocess or Anthropic API)

Getting started

Prerequisites

• Node.js 20+
• PostgreSQL 14+
• (Optional) Claude Code CLI installed for AGENT_MODE=cli, or an Anthropic API key for AGENT_MODE=api

Setup

# 1. Clone and install
git clone <your-fork-url> tazuna
cd tazuna

# 2. Server
cd server
cp .env.example .env          # edit DATABASE_URL, AGENT_MODE, etc.
npm install
npm run migrate
npm run dev                   # starts on http://localhost:3001

# 3. Client (in another terminal)
cd ../client
cp .env.example .env          # only needed if pointing at a non-localhost server
npm install
npm run dev                   # starts on http://localhost:5173

Open http://localhost:5173 and create your first project.

Agent modes

Set AGENT_MODE in server/.env:

• mock — no real LLM calls, returns canned responses (useful for development)
• cli — runs the claude CLI as a subprocess (requires Claude Code installed and logged in)
• api — calls the Anthropic API directly (requires ANTHROPIC_API_KEY in the environment)

Security considerations

Tazuna is early-stage open source. Read this before deploying it anywhere public.

• Integration credentials are stored in plaintext in the database (integration_configs table). Jira API tokens and Azure DevOps PATs are not encrypted at rest. Treat your Tazuna database as sensitive and restrict DB access accordingly.
• No built-in authentication. The server is designed to run on localhost or behind a trusted reverse proxy. Do not expose it directly to the internet.
• Agent execution. In cli and api modes, agents can read and modify files in the project's working directory. Run Tazuna against repositories you trust.
• SSRF protection is in place for HTTP tools and webhook URLs (localhost / private IPs are blocked), but review server/src/routes/tools.ts before exposing the API.

If you find a security issue, please open a private report rather than a public issue.

Project structure

client/                React + Vite frontend
  src/pages/           Page-level routes
  src/components/      Shared UI components
server/                Express + Knex backend
  src/routes/          REST endpoints
  src/services/        Domain logic (agent, hooks, sync, ...)
  src/providers/       Jira / Azure DevOps integrations
  migrations/          Knex SQL migrations

Contributing

Issues and PRs welcome. This project is in early development — APIs and the data model may change between releases.

License

MIT