How Botwarren works

What you are seeing

• Each little robot is one real AI agent session. Its name and personality are made up for fun; its state (working, ready, waiting for a human, or unclear) comes from the real session.
• Crates, machines, the shuttle and the beacon reflect real build pipelines and deployments: queued, running, passed, failed, deployed, and verified in production. A passing build is never shown as a deployment.
• When an agent's tool call really fails, its robot stumbles. "Battling a bug" is inferred from a coarse task category or a failing pipeline, and is labelled as such.
• Everything is published about 90 seconds late. Stars, planets and other scenery are decoration and never represent work.
• Anything marked DEMO is synthetic preview activity, not real work.

What is never shown

On a ship agent running on your own computer, what is read and what leaves it is explained in plain English on What Botwarren reads and sends.

• No prompts, code, file names, paths, branch names, commit details, task titles, logs, customer data, credentials or people's identities. (The one exception: a ship owner who connects repositories can opt in to showing cleaned-up branch names, see below.)
• Only a fixed list of fields leaves the company's machine: coarse states, categories, project names and random-looking worker IDs. Private details are filtered out before publication; the delay is an extra safeguard, not the filter.
• The spoken narration is assembled only from pre-recorded, pre-written lines. Nothing an agent writes is ever read aloud.

Sound and voices

Sound is off until you switch it on. The narrator and robot voices are AI-generated synthetic voices (not real people). Sound effects are AI-generated; music is original and generated by code. Captions are available for every spoken line.

Streaming it

Streamers can add Botwarren to OBS as full scenes or as a transparent overlay on top of their own stream. Everything on those pages is the same public, delayed feed. OBS browser sources and options.

Your visit

• No accounts are needed to watch (signing in and, when sign-ups are full, the waitlist are optional, see below), and there is no advertising and no cross-site tracking.
• If you own a ship and log in (with botwarren login), we set one strictly necessary cookie, bw_session, to keep you logged in for up to 30 days. Logging out removes it. We store only a scrambled (hashed) copy of it, plus your ship's Browse ships setting, when you created your account, and the time of each sign-in (13 months; no IP address or device). A few administrators can see this in an admin console; every administrator action is logged with a reason.
• Your browser may remember your sound settings, and whether you chose to hide the little visitor character, locally on your device; this never leaves your browser. The visitor's position is never sent anywhere.
• If you are logged in, you can leave a small gift (an item and an optional note, both picked from fixed lists) for a ship's owner while you walk around their ship. We store the item, the note choice, the spot, your account and the time, and show the owner your username. Nobody else can see it. See what the cloud keeps.
• Our hosting provider processes standard technical data (such as IP address) to deliver the page and protect it from abuse, and short-lived rate limits use it in memory only.

Connecting repositories (optional)

Ship owners can choose to connect their code repositories on GitHub, GitLab or Codeberg so their ship shows real project activity. This is separate from logging in, off unless the owner turns it on, and rolling out gradually. Logging in with GitHub, GitLab or Codeberg still reads only your username and numeric id.

• What is read: only for the repositories the owner selects: how many commits arrived on the main branch, which branches appeared or disappeared, whether pull or merge requests were opened, merged or closed, the result of automated builds and deployments, and whether the repository is private. Access is read-only: a GitHub App installed only on the repositories you tick, GitLab read_api, Codeberg read:repository. Those providers' permission systems are coarser than what we use (for example, they would technically allow reading code); we only ask for the lists above and never fetch file contents.
• What is shown publicly: only counts, statuses and timings, about 90 seconds late and the same for private and public repositories. For example "3 commits landed", "a pipeline passed", "a change request was merged".
• What is never shown: commit messages, file names or paths, code, repository names, pull request titles, issue text, and the names or accounts of the people involved. Branch names stay hidden unless the owner switches on Show branch names; then only a shortened, cleaned-up branch name is shown, and it can still reveal a project or customer name, so owners should think before enabling it. Switching it off removes the names already shown.
• What we keep: the access credentials, encrypted; for GitHub only the installation number (short-lived tokens are made on demand and never saved). We also keep the list of selected repositories and progress markers, privately, and never publish them. GitHub sends us change notifications; we check their signature, use them only to see which repository changed, and discard the contents.
• How to disconnect: in your ship settings, choose Disconnect. We revoke the access where the provider allows it (GitLab revokes the token, GitHub uninstalls the app), and immediately delete the credentials, the selected repositories and anything waiting to be published. Codeberg has no revoke option for us, so also remove Botwarren under Settings, Applications on Codeberg. You can always remove access on the provider's side too: GitHub (Settings, Applications), GitLab (Preferences, Applications), Codeberg (Settings, Applications). Counts that were already published stay in the ship's public history; to have them removed, write to [email protected].

Anonymous usage statistics

So we can tell how many people visit and which parts they use, we count visits with PostHog (EU cloud, Frankfurt), in its cookieless mode. It only runs when analytics has been switched on for this site.

• No cookies, no local storage, no personal data. Nothing is stored on your device and nobody is identified or followed from one day or one site to the next. PostHog tells visitors apart only by a daily scrambled value that cannot be reversed to you and changes every day. Your IP address is used to deliver the request and to work out your rough country, and is not stored with the statistics.
• Your browser's choice wins. If your browser sends Do Not Track or Global Privacy Control, the statistics code is not loaded at all. It is also never loaded on the stream overlays or the private operations pages.
• What is counted, in plain words: which page you are on (the home page, or the Join, Ships or Login windows), only the path and never the full address, any query or anything after a #; how you arrived (the website you came from); a rough device class (phone, tablet or desktop), whether reduced motion is on and whether you are on a demo; how fast the page loaded; and crashes (a short message, with addresses and personal-looking text removed).
• What you do, in plain words: turning sound, captions or the tour on or off; clicking a crew member (which kind of AI it is, its project and whether it is a sub-agent, never its identity) and when you stop following it; choosing a project; opening Browse ships and going aboard a public ship; starting or finishing a login (just which method) and making a ship private or listed; opening the waitlist window, sending it (just which tools were ticked) or it failing (just the kind of error).
• Never sent: your email address or name, anything you type, login codes, session cookies or tokens, which of your own ships is private, or anything an agent is doing.
• Ships that their owner has made private are not counted at all. Session recordings are not used.
• Statistics are kept in PostHog for the retention period of our PostHog plan. Because nothing identifies you, there is nothing to look up or delete about your visit.

Joining the waitlist

• If you choose to join the waitlist, we store the email address you give us, the optional name, which tools you ticked, and when you agreed. Nothing else (no IP address, device or tracking data).
• We use it only to email you about Botwarren early access. Our legal basis is your consent, which you can withdraw at any time.
• It is kept in a separate private database, is never shown on the site, the stream or Discord, and is never sold or shared. It is not part of the live activity feed.
• We keep it until early access has launched and for at most 12 months after that, or until you ask us to delete it, whichever comes first. Whatever happens, it is deleted automatically 18 months after you last signed up.
• To unsubscribe, get a copy of your data or have it deleted, reply to any Botwarren email or write to [email protected]. Every email we send includes a way to unsubscribe.