Why Devuntu

• AI agents as assignees

  Assign a ticket to Claude Code / Codex / OpenCode and, once an approver picks the agent mode (plan first or auto), it gets processed, even while you sleep. Plans and reports stay on the ticket, where humans can approve them or send them back

• Built-in MCP server

  Create, update and report on tickets from the AI tools you already use, with your own permissions. Per-board premises (repository, conventions) reach the AI too

• Self-host with one compose.yaml

  Runs with Docker Compose and keeps your data on your own server. MIT licensed, with no seat limits or fees

Features

• Kanban boards and tickets

  Boards, tags, assignees, priorities, due dates, comments and mentions. Parent/child and related tickets, acceptance criteria, per-board ticket templates and change history. File attachments in descriptions and comments, and Markdown with mermaid diagrams, syntax-highlighted code, alerts and footnotes. Filter down to the children or related tickets of a ticket, and export tickets or plans / reports as PDF or HTML

• Calendar

  Google Calendar integration, and sharing your free/busy time through a public URL that needs no sign-in

• Notifications

  Email / Slack DM / web push notifications for mentions, assignee changes, due date reminders and AI agent run results. Team boards can also post to a Slack channel, including announcements such as release news

• MCP server

  AI agents can work on tickets with your own permissions

• AI agents

  Register AI agents as assignees and let them process their tickets automatically (Claude / Codex / OpenCode)

• Task splitting

  Agents can split a large ticket into child tickets; once you approve the proposal, they work through the children in order

• AI context

  Per-board premises such as the target repository and conventions, delivered to MCP clients and agents

• Knowledge / memory

  Per-board knowledge curated by users and memory recorded by AI while working, shared by members and AI

• GitHub / GitLab

  Link branches, pull requests and commits to tickets and see the PR status and CI results. Complete tickets on merge (held while acceptance criteria are unmet), and send them back to the agent on CI failures or review comments

• Remote execution

  Run predefined jobs on remote servers and follow the output in real time

• Authentication

  Email OTP, passkeys and Google sign-in. Devuntu can also act as an OAuth provider

• English / Japanese

  Switch the UI language; notifications are sent in the language you choose

Screenshots

Get started

Devuntu is distributed as a Docker image and runs with Docker Compose. The only file you need is compose.yaml.

Requirements: Docker Compose v2.24+ / Docker Engine 25.0+, 2GB of memory or more (4GB recommended), and a way to send email for sign-in. A reverse proxy if you serve it over HTTPS

1. Place compose.yaml on the host (curl -fsSLO https://raw.githubusercontent.com/playree/devuntu/stable/compose.yaml)
2. Generate the configuration files with docker compose run --rm tools setup-env
3. Start it with docker compose up -d --wait (database migrations run on startup)
4. Open <BETTER_AUTH_URL>/start and register the first administrator

See the quick start in the README for details. The application UI is available in English and Japanese.

Documentation

• For users

  The first 30 minutes, how to use each screen, connecting AI tools and handing tickets to agents

• For operators and administrators

  Installation, HTTPS, Slack / GitHub / GitLab integrations, backup and operations, admin screens

• For developers

  Development environment, contributing, and how each feature works internally

The detailed documentation is available in Japanese only. Please use a translation tool or an AI assistant.