Quick start · How a card moves · Local or Cloud · Officium Cloud

Your agents can write code all day. What they can't do is decide what matters, notice that a task is half done, or know when to stop and ask you. Officium is the board where that gets settled: agents pick their next card over MCP, move it as they work, prove it at review, and hand back what only you can decide. You see all of it as it happens.

_{An agent moves SHOP-3 over the API while the board is open. No reload: the card travels to Review, and a reviewer sends it back with its reason.}

This repository is Officium Local: free, self-hosted, for one person on one machine. Officium Cloud is the hosted edition for teams.

Quick start

You need Docker.

git clone https://github.com/CMolG/officium.git && cd officium
docker compose up -d --build
scripts/demo-seed.sh            # optional: a demo project to look around

Open the board at http://127.0.0.1:3190, then connect your agent:

claude mcp add --transport http officium http://127.0.0.1:8190/officium/mcp

Ask it "take the next Officium card". It discovers the operations with discover and runs them with query: cards.next, cards.move, cards.search and the rest. Any other agent that speaks MCP over HTTP connects to the same URL.

Everything listens on 127.0.0.1 only. To require a token, set OFFICIUM_LOCAL_TOKEN before docker compose up; agents then send Authorization: Bearer <token>.

How a card moves

flowchart TB
  you(["You"])
  agent(["Your agent<br/>Claude Code, Codex..."])
  subgraph officium ["Officium"]
    direction LR
    todo["To do"] --> doing["In progress"] --> review["Review"] --> done["Done"]
    review -- "sent back, with reasons" --> doing
    doing -- "only a person can decide" --> needs["Needs you"]
  end
  you -- "writes and orders the cards" --> todo
  agent -- "cards.next, cards.move" --> doing
  needs -. "you answer" .-> you

Loading

1. You write the card and say how it counts as done. Transitions carry contracts, so a card can't reach Review without what Review needs.
2. An agent asks for work. cards.next hands it the most important card it can start, skipping blocked ones.
3. It works, and the board shows it. Every move is on your screen the moment it happens.
4. Review says why. A card sent back carries its reasons from a closed list (tests missing, acceptance unmet, plan not followed...), so the next attempt knows exactly what to fix.
5. What only you can decide waits for you in the Needs you lane, not in a chat scrollback.

What you get

• One board for people and agents. Projects, cards, epics and sprints, with a status workflow whose transitions carry contracts.
• Work an agent can pick up on its own. Priorities and blockers decide what cards.next answers, so you order the work once and agents follow it.
• Epics as checkpoints. An agent can sweep a whole epic in one session, checkpoint by checkpoint, so long work stays inside what models do reliably.
• Review with reasons. Send-backs use a closed list of reasons, so you can see where work keeps failing.
• Needs you. Cards only a person can resolve wait in their own lane until you answer.
• Roles. A role matrix decides which agent role refines, builds and reviews each kind of card.
• Docs next to the work. A documentation tree with pages, search and history.

Local or Cloud

	Local	Cloud
Board, cards, epics, sprints, workflow contracts	✓	✓
MCP server for agents (cards.next, review send-backs, Needs you)	✓	✓
Epic checkpoints, role matrix, docs	✓	✓
Runs on your machine, your data stays there	✓
Teams: organizations, members, invitations and sharing		✓
One account across devices, nothing to host or back up		✓
Real-time presence and collaborative editing of docs		✓
Email notifications, and forms to answer Needs you cards		✓
GitHub integration and a CI integration gate		✓
Changelog and test history from your pipeline		✓
Usage, cost and time reports		✓
Queue cards from the board straight into a running agent window		✓

Cloud features appear in Local with an Officium Cloud label, so you can see what they are. Nothing in Local nags you to upgrade.

Try Officium Cloud →

Questions

Is it free? Yes. Officium Local is free to use, modify and self-host for your own work and your company's, under the Elastic License 2.0. The one thing it does not allow is offering Officium to others as a hosted service.

Does anything leave my machine? No. Officium Local sends no telemetry, has no account to sign in to, and stores everything in the PostgreSQL container on your computer.

Which agents work with it? Any agent that speaks MCP over HTTP. It is built and used every day with Claude Code.

Can I move a Local project to Cloud? Not yet: there is no import between the two editions today. Cloud is the same board with team features on top.

Officium and Fluxor

Officium tracks the work. Fluxor gives your agents the rules, skills and hooks they work with, including autoloop, which runs several agents in parallel on Officium's backlog, each in its own git worktree.

Background

Officium is the working implementation of FIERCE, a framework for software development where agents carry the implementation and a few experts hold the decisions:

Molina Galindo, C. (2026). FIERCE: Fast Implementation Ensuring Results by Capable Experts. doi:10.5281/zenodo.23972303

Repository layout

• server/: the API (Java, Quarkus): REST under /officium/v1 and MCP under /officium/mcp.
• web/: the board (Next.js), built in its Local edition.
• docker-compose.yml: PostgreSQL, the API and the web, all bound to 127.0.0.1.
• scripts/demo-seed.sh: a demo project. scripts/smoke.sh: an end-to-end check.

This repository is published from the private monorepo where Officium is developed. Issues are welcome; pull requests are ported by hand.

License

Elastic License 2.0. See NOTICE.