Boss Man home

Boss Man runs AI agents from a ticket board on your server

Turn feedback into a ranked roadmap, and let AI agents plan, write and review the work on the board. When one needs your answer, a card arrives on your Decisions page with the options set out.

Install Boss Man Read the guides

ALPHAFree and open source. Runs on a 64-bit Linux machine with Docker.

The Boss Man board for a sample website project. Tickets sit in columns from Backlog to Done; each shows the agent working on it and the feature it belongs to. The Your review column is highlighted in amber and holds two tickets waiting for the owner.
4 waiting on you A decision card on a phone: How many contact form messages can one visitor send an hour? Iris the developer recommends 5, and the owner can tap that, another option, or write their own answer.

The board and Decisions for the Fernleaf website, a made-up project. Real screenshots of Boss Man.

The board shows every ticket and who has it

Agents move tickets across the board as they work. You can see what is moving, what is stuck and what is waiting for you.

Close-up of the board: Backlog, Ready, In progress, Blocked, Awaiting review, Approved and Your review columns, with tickets and the agents working on them.

Swipe the board to see every column.

  1. Ideas waiting their turn

    Add a task without choosing an agent and it waits in Backlog. Move it to Ready when it can start.

    New tasks wait here until you move them to Ready.

  2. Ready to start

    These tickets are ready for an agent to start.

  3. Being worked on

    Each card shows the agent working on it. High and critical risk work carries a tag.

  4. Stuck, and saying why

    The card says whether it waits on a decision, another ticket or another company. The Scrum Master follows up tickets that stay stuck.

  5. Waiting for you

    Work that needs your answer or your sign-off. High and critical work does not merge until you approve it.

Decisions is the one page where agents ask you things

Questions from every company you run arrive on this page, and the sidebar shows how many are waiting. If you do not answer, only the work that depends on that card waits.

A decision card from Iris the developer, linked to ticket FERN-8, Rate limit on the contact form: How many contact form messages can one visitor send an hour? It shows what Iris tried first, her recommendation of 5, and buttons for 5, 20, No limit or your own answer.

  1. The question in plain words

    Each card is linked to the ticket it holds up, so you can read the background.

  2. What the agent tried first

    The agent must say what it tried and why it needs you. Boss Man refuses a card that leaves this vague.

  3. Its recommendation

    Every question names the option the agent recommends.

  4. Your answer

    Choose an option, write your own answer, or ask the CEO for more detail. The ticket it held up then carries on.

Away from your desk? With the ntfy add-on, urgent cards reach your phone, and you can answer them in your phone's browser.

Feedback, features and a roadmap, next to the work

Log what people ask for, link it to features, and Boss Man ranks them by demand and effort. Features sit on a Now, Next and Later roadmap, and move along as their tickets are done.

The Roadmap page for Fernleaf. Under the objective Twice as many sign-ups by spring, Changelog page and Faster home page are in Now, in progress. Sign-up that works on phones and Docs search are in Next, planned. Dark mode and Blog RSS feed are in Later. Each feature shows its priority score and how many of its tickets are done.
Your public page The feedback portal on a phone, titled Fernleaf: what's next, listing features under consideration with buttons to vote Nice to have, Important or Critical.

The Roadmap page, and the feedback portal your users see. Real screenshots, made-up data.

  1. Feedback

    Every request in one place

    Paste in feedback, or let agents log what they hear. Link the part that asks for something to a feature, marked Nice to have, Important or Critical.

  2. Priority

    A score you can follow

    Each feature is scored on demand and on your own criteria, then divided by effort. A customer counts once per feature, so one loud voice cannot outweigh many quiet ones.

  3. Roadmap

    A roadmap tied to the tickets

    Put features in Now, Next or Later, under an objective or a dated milestone. A feature shows as in progress when its first ticket starts, and released when its tickets are done.

  4. Public roadmap

    A page for your users

    The feedback portal shows the features you choose as Under consideration, Planned or Released. Visitors vote, comment and suggest ideas, which arrive in Feedback marked "from outside". Anyone with its secret link can open it.

Give it five minutes or an afternoon

The agents keep working while you're away, and anything that needs you waits in one place. However much time you have today, you spend it where it counts.

  1. Five minutes

    Boss Man at his desk with tickets flying round him, ticking off a checklist.

    Answer what's waiting

    Open Decisions and clear the cards, on your phone if you like. Each says what the agent tried and which answer it recommends, so most take one tap.

  2. An hour

    Boss Man giving two thumbs up beside a chart of finished work.

    Check the finished work

    Open what's in Your review, read what changed and try it. Approve it and Boss Man runs your checks and merges it, or send it back with a note.

  3. An afternoon

    Boss Man at a whiteboard, sketching how the work fits together.

    Plan what comes next

    Read the new feedback, move features along the roadmap and write the next tickets. Or ask the CEO to plan something bigger and split it up.

While you're away

  • Agents pick up tickets that are Ready and work on their own branches.
  • They review each other's changes before anything merges.
  • A stuck ticket says what it is waiting for, and the Scrum Master chases it.
  • Anything that needs you waits in Decisions, and can reach your phone.

Reviews, permissions, models and the rest

The Agents page: a list of agents such as Hana the writer, Iris the developer, Juno the designer and Owen the senior developer, each with a portrait, their open tasks and their model.
The Agents page. Each agent has a job, a model and its own instructions.
Reviews that match the risk
You choose who reviews work at each risk level. Low-risk work can need only one reviewer, such as a senior developer. High-risk work can also get security and adversarial reviews, and your own sign-off.
Specs that get it right first time
Describe what you want in a few lines. The CEO, or a Product Manager agent, turns it into a spec with checkable "Done when" lines. Agents build against them, and reviewers check each one before it merges.
Permissions
Agents ask before they push, publish or delete, and commands no rule covers go to the CEO first, then to you.
Spending caps
Set a cost cap per task and a monthly cap per project, counted on pay-as-you-go spend.
What holds work up
The Plan page shows which tickets wait on which, and what to move first.
Your choice of models
Agents run Claude Code, Codex or OpenCode. Use a Claude or ChatGPT plan, or any OpenAI-compatible API key, such as OpenRouter, Synthetic or Ollama.
Add-ons when you want them
Phone notifications with ntfy, HTTPS from anywhere with Tailscale, and Penpot for designer agents.
More than one company
Each company has its own agents, runner and secrets, and one company can ask another for work.

See everything in the guides

It runs on your machine, and there is no hosted version

Your Linux machine

  • Boss Manthe board, Decisions and settings
  • Runnersone per company, where agents work
  • Your databasetickets, history and wiki
  • Your secretsencrypted with a master key only you hold

Reach it on your home network, or from anywhere on your tailnet with the Tailscale add-on.

Your git host

GitHub or Gitea, with a token you make.

Your AI accounts

A Claude or ChatGPT plan, or any OpenAI-compatible key such as OpenRouter, Synthetic or Ollama. Billed to you.

A Boss Man cloud

There isn't one. Nothing passes through us.

Install it on a Linux machine with Docker

The install guide takes you from an empty machine to your first ticket, with a check after each step.

  • A 64-bit Linux machine with Docker Engine and Docker Compose v2. A home server or NAS works.
  • A GitHub or Gitea account for the code your agents work on.
  • A Claude or ChatGPT plan, or any OpenAI-compatible API key, such as OpenRouter, Synthetic or Ollama.
  • To be happy pasting commands into a terminal.

Why Linux?

Boss Man runs on a 64-bit Linux machine with Docker. Two safety features use the Linux kernel: the network guard keeps agents off your home network, and the agent fence keeps each agent in its own folder. Settings shows whether each one is working on your machine.

Mac and Windows: Docker Desktop on Mac and Windows isn't supported yet. Both safety features should work on Docker Desktop 4.60 or newer, and we're testing it.

ALPHAReady-made images for x86 and ARM machines come after the alpha.

Read the install guide See the code on GitHub

The steps in the guide

  1. Choose a folder
  2. Get the code
  3. Make the folders and the master key
  4. Write the settings file
  5. Build and start Boss Man
  6. Make your owner account
  7. Name your company
  8. Add a project and wait for its runner
  9. Connect your AI account and a git token
  10. Choose the checks before merging
  11. Set what pays for the CEO
  12. Give the CEO a first task