Squad33 is a kanban board that runs on your own computer and hands your tickets to a coding agent — Claude Code, or another agent you configure — that also runs on your own computer. By installing or using Squad33 you accept the terms below. They are written to be read, not skimmed: each one describes something Squad33 actually does and what part of it is yours to own.

1. The coding agent runs with your real permissions

When Squad33 works a ticket or answers you on the voice screen, it runs a coding agent as you, on your machine. The agent can create, modify and delete files, run programs, and use the network — the same things you can do from a terminal. That is what makes it useful, and it is also what makes it able to do real damage if a ticket or a spoken instruction is incorrect, ambiguous, or malicious (for example, instructions pasted from a source you did not vet).

You are responsible for what you ask the agent to do and for reviewing what it produces before you rely on it. Squad33 is the board and the plumbing; the changes are made under your account, at your request.

2. A paired phone is a door to this computer

Pairing a phone opens a tunnel from that phone to the computer Squad33 runs on, through Squad33's relay, so the board and the voice screen work away from your desk. While the tunnel is open, anyone holding a paired phone can reach this computer through it — including whoever holds the phone after it is lost or stolen.

Keeping that door closed is your responsibility, and Squad33 is not responsible for what comes through it while it is open. The tunnel switch in the top bar closes it when you are not using it; if a paired device is lost, close the tunnel and revoke that device (Admin → Phone Pairing) promptly.

3. Voice conversations are stored on this computer

To draw the conversation on the voice screen — and to redraw it after a reload, or on your phone — Squad33 keeps the full transcript of both sides of every voice conversation in its local database, in a table called voice_turns. That transcript is stored as spoken, not redacted, and it stays until you clear the conversation from the voice screen. The database lives in Squad33's data directory on this computer; nothing in it is sent to Squad33.

The Mac app's diagnostic logs are different: since they exist to be shared when something breaks, the packaged app replaces conversation content in them with character counts before it reaches disk. The transcript in the database is not covered by that redaction — it is the conversation, kept so the screen can show it.

If you select the hosted transcription engine and provide an OpenAI API key, recordings of your speech are sent to OpenAI to be transcribed, under OpenAI's own terms. The default engine transcribes on this computer.

4. The agent acts through access you already have

Squad33 does not ship, hold, or read your credentials. It has no copy of your AWS keys, your GitHub token, or your Claude account — the agent works through whatever access already exists on this machine, such as a gh CLI login, an SSH key, or the Claude Code account you are signed into, and its usage is billed to those accounts under their own terms.

That also means the agent can do whatever that existing access permits. You are responsible for what the accounts and credentials on your machine allow the agent to reach.

5. Crashes are reported automatically

When one of Squad33's own background processes hits an error it does not know how to handle, the packaged Mac app sends a crash report to Squad33 by itself, without asking. That is how bugs get found in an app nobody is watching, and accepting these terms is what turns it on.

A crash report holds the error's type, its message, the Python traceback, which of the three processes it came from, the app version, a timestamp, and this install's id — the same short random id the tunnel routes on. That id is the only thing in it that identifies anything, and it identifies an install: no name, no email address, no IP address, no computer name, no account name of yours. File paths in a traceback have your home directory replaced with ~ before it is sent, and the error's text is scrubbed of anything shaped like a password, an API key or an email address. Nothing you said, typed or asked for is in it, and no local variables are collected.

You can turn this off at any time, under Admin → Log, and nothing is sent after that. Repeats are not sent either way: the same error is reported at most once every few minutes, and a checkout of the source code — as opposed to the Mac app — never reports at all.

6. Beta software, provided as-is

Squad33 is beta software. It is provided "as is" and "as available", with no warranty of any kind — no promise of uptime, availability, error-free operation, or fitness for a particular purpose. Features may change or break, and to the maximum extent permitted by law, Squad33 LLC is not liable for damages arising from your use of it, including work lost to an agent's mistake or exposure through a tunnel left open. Back up anything you cannot afford to lose before pointing an agent at it.

---

Questions about any of this: hey@squad33.ai.