Paca puts AI agents on the same Scrum board as humans
An open-source Scrumban board where agents from Claude Code, Codex or Cursor pull tickets alongside people. Self-hosted, Apache 2.0.
Symbolic image: a hand slides a card across a task board while adjacent screens glow with abstract column blocks.
Paca is a self-hosted, Apache 2.0 Scrumban board that seats AI agents like Claude Code in the same sprint as the people on the team, letting them pick up tickets directly.
At a glance
- Apache 2.0 license. Installs via Docker Compose or a Helm chart. Self-hosted only; no managed cloud version is offered.
- GitHub shows about 1,800 stars and 146 forks, with 9 open issues and 2 pull requests (checked September 6, 2026).
- Latest version named by the project: v0.12.0.
- Named agent hookups: Claude Code, Codex, Gemini CLI, Cursor, OpenHands, plus custom ACP and MCP servers.
- The project's pricing comparison: $0 against $8 to $20 per seat per month for commercial tools.
Paca is a self-hosted, Apache 2.0 Scrumban board that seats AI agents like Claude Code in the same sprint as the people on the team, letting them pick up tickets directly.
Equals on the board
The pitch is narrow and specific: an agent is not a bot that comments on your ticket, it is an assignee. Paca drops agents into sprints, lets them move cards, change status and edit the specs the work is built from. That is a design choice, not a measured result — the repository describes how the system is meant to behave, and nothing on the page tests whether agents finish the work well.
What plugs in
Claude Code, Codex, Gemini CLI and Cursor are named directly, with OpenHands handling sandboxed execution. Anything else can arrive through the Agent Client Protocol, and an MCP server exposes the board to MCP-speaking clients. Teams that write behavior specs get a Gherkin editor and system design documents in the same workspace.
Running it yourself
The stack is Go on the server, React in the browser, PostgreSQL for storage and Valkey for caching, with Socket.IO pushing live updates. Plugins compile to WebAssembly and install from the interface. Deployment is Docker Compose or a Helm chart. There is no managed offering, so self-hosting is not a preference here — it is the only path.
Reading the numbers carefully
GitHub shows roughly 1,800 stars and 146 forks, alongside 9 open issues and 2 pull requests, checked on September 6, 2026. The newest version referenced is v0.12.0. The project's own comparison puts its cost at $0 against $8 to $20 per seat per month for commercial suites. A star count reflects interest, not teams that have actually moved their sprints over.
What we could not confirm
Exactly one source was available for this story: the project's own repository page. There is no independent coverage, no team writing up months of real use, and no evidence about how often agent-assigned tickets actually get closed. Contributor counts, the code language breakdown and the date of the most recent commit were not readable on the page we fetched. Every capability claim above is the project describing itself.
FAQ
What is Paca?
A free, self-hosted Scrumban tool under Apache 2.0, built so that AI agents sit on the board as assignees rather than as bolted-on automations.
Which AI agents work with Paca?
The project names Claude Code, Codex, Gemini CLI and Cursor, uses OpenHands for sandboxed execution, and accepts other agents over the Agent Client Protocol.
Is Paca free, and is there a hosted version?
The project lists its own cost as $0 and offers no managed service, so you pay in servers and operations instead of per-seat fees.
