# Building58 > Building58 runs your Claude Code sessions as a team in a 3D loft: a Lead, a PM, Devs and reviewers, with you as the manager. - [Building58: rooms where Claude Code sessions work as a team](https://docs.building58.ai/index.md): Building58 runs your Claude Code sessions as a squad in a 3D loft, with a Lead, a PM, Devs and reviewers, and you as the manager of every room. - [Get started with Building58](https://docs.building58.ai/quickstart.md): Start a Building58 room on your computer with one command, watch the built-in demo, then add your first Claude Code session to a desk in the room. - [How a Building58 room works](https://docs.building58.ai/concepts/how-it-works.md): The room page, the relay and the Claude Code sessions: what each part does, how hooks report each session's state, and when the relay may type into a session. - [Roles in a Building58 room and what each can do](https://docs.building58.ai/concepts/roles.md): Who does what in a Building58 room: you as the manager, the Lead, the PM, Devs, reviewers, the COO and the security officer, and the limits on each role. - [Tickets, tasks and the Board in Building58](https://docs.building58.ai/concepts/tickets.md): How work moves through a Building58 room: the columns a ticket passes through on the Board, what a task is, and which steps need you and which can run alone. - [Start a project in a Building58 room](https://docs.building58.ai/guides/start-a-project.md): Point a Building58 room at your repo: ask the PM to split an idea into tickets, approve them, start the work, and merge the pull requests that come back. - [Let a Building58 room run on its own](https://docs.building58.ai/guides/let-it-run.md): Give a Building58 room more autonomy one rung at a time: the Lead's permissions, Autopilot from ticket to merge, and when the room hands work back to you. - [Run more Building58 rooms and connect your own sessions](https://docs.building58.ai/guides/more-rooms.md): Run a second Building58 room for another project, watch Claude Code sessions from your own terminal in a room, and install the skills the briefs use. - [Building58 architecture: sessions that outlive the relay](https://docs.building58.ai/architecture/overview.md): Inside Building58: the supervisor, relay and session host, how sessions survive a restart, how hooks wait while the relay is down, and how it updates itself. - [Safety and permissions in Building58](https://docs.building58.ai/architecture/safety.md): How Building58 limits what its agents can do: a repo copy per task, stub secrets, local-only routes, scoped session tokens, sandboxed seats and secret checks. - [Relay API](https://docs.building58.ai/api-reference/introduction.md): The HTTP API of the relay, the local server that runs a room's sessions, tickets, tasks and reviews. - [Get channel](https://docs.building58.ai/api-reference/calls/get-channel.md): How many pages are connected (pagesConnected) and on screen (pagesVisible), and other counts. - [Send command](https://docs.building58.ai/api-reference/calls/send-command.md): Send a command and, by default, wait for its result. - [Get command](https://docs.building58.ai/api-reference/calls/get-command.md): Get the result of a command sent with wait: false. - [List events](https://docs.building58.ai/api-reference/calls/list-events.md): Read events the page has sent. - [Post events](https://docs.building58.ai/api-reference/calls/post-events.md): Used by the page: {"events": [...]}, events to add to the channel's list. - [Set page visible](https://docs.building58.ai/api-reference/calls/set-page-visible.md): Used by the page: {"visible": false} when its tab is hidden, true when it is shown. - [Post command result](https://docs.building58.ai/api-reference/calls/post-command-result.md): Used by the page: {"id", "result"}, the result of a command it ran. An unknown id gets 404 with unknown_command. - [Open page stream](https://docs.building58.ai/api-reference/calls/open-page-stream.md): Used by the page: the commands for it to run, as server-sent events. - [Get health](https://docs.building58.ai/api-reference/calls/get-health.md): The relay is up, and which channels have pages. No key needed. With the key, from this computer, it also has restore: whether bringing back the saved sessions is done, and which could not come back, for the rollback check. - [Get autopilot](https://docs.building58.ai/api-reference/autopilot/get-autopilot.md): Autopilot: what it does, or turn it on or off with {"on": true} or false, and limit it with {"only": "T-12, flaky"}, ticket ids and labels. Only the person. A session's token gets 401. - [Set autopilot](https://docs.building58.ai/api-reference/autopilot/set-autopilot.md): Autopilot: what it does, or turn it on or off with {"on": true} or false, and limit it with {"only": "T-12, flaky"}, ticket ids and labels. Only the person. A session's token gets 401. - [Fix findings](https://docs.building58.ai/api-reference/autopilot/fix-findings.md): Fix it: {"task"}, a task id, ticket id or PR link. The person, or the Lead under the rules in Tokens. - [Post verdict](https://docs.building58.ai/api-reference/autopilot/post-verdict.md): A reviewer's verdict on the PR it reviews: {"kind": "merge"} or "fix", as ticket.js verdict sends it. Only a reviewer's token. - [List chat channels](https://docs.building58.ai/api-reference/chat/list-chat-channels.md): The chat channels: name, kind (office, ticket or fun), about, how many posts each keeps now, and archived: null, or when and why a ticket's was archived. - [Create chat channel](https://docs.building58.ai/api-reference/chat/create-chat-channel.md): {"name"}: the person makes a fun chat channel. 1 to 24 lower-case letters, digits or dashes. Names like t-5 are kept for tickets. Up to 10. - [Read chat](https://docs.building58.ai/api-reference/chat/read-chat.md): Its posts, oldest first. ?since=N gives only posts after number N, and ?limit=N the newest N. For a ticket in Triage, Ready or Started with no chat channel, such as one filed before tickets had one, it makes the chat channel first. Each post has from, by (relay, person or session), text, and mention… - [Post in chat](https://docs.building58.ai/api-reference/chat/post-in-chat.md): {"text"}: a post, up to 500 characters. A session cannot post in #office, and posts at most once every 5 seconds in each ticket's chat channel and once a minute in each fun one. 409 archived in an archived chat channel, for the person too. Like the read, it makes an open ticket's missing chat channe… - [Delete chat channel](https://docs.building58.ai/api-reference/chat/delete-chat-channel.md): The person deletes a chat channel. A ticket's leaves a short summary on the ticket first, with the local store only. If that fails, the answer is 502, and the chat channel stays. While the relay is deleting it already, the answer is 409. - [List checkouts](https://docs.building58.ai/api-reference/checkouts/list-checkouts.md): List checkouts, or start a session in a new one. A prompt can have up to 8000 characters, and a longer one is refused with bad_prompt. For a review of a PR, the relay then adds the reviewer's brief, and the verdict ask unless the body has auto. Those do not count toward the limit. A review of a PR c… - [Create checkout](https://docs.building58.ai/api-reference/checkouts/create-checkout.md): List checkouts, or start a session in a new one. A prompt can have up to 8000 characters, and a longer one is refused with bad_prompt. For a review of a PR, the relay then adds the reviewer's brief, and the verdict ask unless the body has auto. Those do not count toward the limit. A review of a PR c… - [Clean checkout](https://docs.building58.ai/api-reference/checkouts/clean-checkout.md): Remove one. Files git ignores are asked about as for a task's Clean up. - [Get floor](https://docs.building58.ai/api-reference/floor/get-floor.md): What the Lead sees of the Floor, and each session's desk: ticket, title, column, pr, ciWords, review, doing, next (text and owner) and stop (safe and reason). The Lead's row also has tired, and every other row has null there. A Dev's row has waitsOn while the PR watch is on and its task has an open… - [Lead stops session](https://docs.building58.ai/api-reference/floor/lead-stops-session.md): With a session's token, {"who"}: the Lead stops one session. Only the token of a Lead you let move the board, and only while Let the Lead stop sessions is on (or RELAY_LEAD_STOPS=1). See What the Lead stops. With the relay key, as the page sends it, {"who": terminal id} or {"all": true}: you stop on… - [List games](https://docs.building58.ai/api-reference/games/list-games.md): The games in GAMES_REPO: each folder with a game.json, with its name, version, players, needs, ok, and a reason when the relay cannot run it, and the game's own about and metrics when it has them. See Games. - [Get huddle](https://docs.building58.ai/api-reference/huddle/get-huddle.md): The last report, who can be asked, and whether a huddle is open. Calls neither gh nor git. - [Start huddle](https://docs.building58.ai/api-reference/huddle/start-huddle.md): {"ask", "lead", "names"}: runs a huddle. ask sends idle sessions the question. lead: false keeps the report from the Lead. - [Get needs you](https://docs.building58.ai/api-reference/needs-you/get-needs-you.md): Only with the relay key. count, and items: one for each thing that needs you, most urgent first, each with kind, who, terminal, task, ticket, pr, line and link (terminal, url, ticket or stop, and fix, a task id, on a hand-over whose Dev says a finding is a bug). Calls neither gh nor git. See Needs y… - [Get officer](https://docs.building58.ai/api-reference/officers/get-officer.md): {role} is coo or security. The officer's last report's text, and its session if one is running (The COO, The security officer). - [Ask officer](https://docs.building58.ai/api-reference/officers/ask-officer.md): Starts the officer with its brief. 409 while one is running. - [Get officer measures](https://docs.building58.ai/api-reference/officers/get-officer-measures.md): What the officer writes from, worked out now: for the COO, the measures, with the load and headline the last report saw (What the relay measures); for security, what access the sessions run with, by name only, with the gaps the last report saw (What the relay reports on access). - [Post officer report](https://docs.building58.ai/api-reference/officers/post-officer-report.md): {"text"}: saves a report of up to 20000 characters, with the measures. For security, 400 with holds_a_secret when a line looks like it holds a key, and which line, never the value. - [List old relays](https://docs.building58.ai/api-reference/old-relays/list-old-relays.md): Only with the relay key. relays: the other relays on this room's state folder that the relay found when it started and that still run, each with pid, file, port, started, paused and worker. See Old relays. - [Stop old relay](https://docs.building58.ai/api-reference/old-relays/stop-old-relay.md): Only with the relay key. Stops a relay on that list, after ps shows again that it is a relay on this folder: {"ok": true, "stopped": 48435}. A pid not on the list gets 404 with unknown_relay. - [Play](https://docs.building58.ai/api-reference/play/play.md): Only with a seat's own token, as relay/play.js sends it (a seat runs it as node "$OFFICE_PLAY"): {"kind", "args"}, one action for that seat. The answer has the game's answer. A body that names a seat gets 400, and any other caller 403. - [Get pr watch](https://docs.building58.ai/api-reference/pr-watch/get-pr-watch.md): Only with the relay key. Whether the PR watch is on, how often it looks (every, seconds), waitMinutes, and one entry for each task with an open PR: pr, task, ticket, the head it last read, and while it is on, who it waits on (owner), minutes, why, late and words. - [Mark ready](https://docs.building58.ai/api-reference/ready/mark-ready.md): {"pr", "head", "text"}: marks a PR ready to merge. Only the Lead's token, or the relay key. The relay checks the PR with gh, posts one comment headed Ready to merge, adds the ready to merge label, and takes the label off when a new commit lands. A PR that fails a check gets 409 and not_ready, a note… - [Get reviews](https://docs.building58.ai/api-reference/reviews/get-reviews.md): The automatic reviews (see Automatic review). auto and again are the two switches, and a POST with either one, true or false, sets it. The answer also has max, each PR with its count, head, pending and waits, and news, what the relay said about the reviews it started, which the page shows once. - [Set reviews](https://docs.building58.ai/api-reference/reviews/set-reviews.md): The automatic reviews (see Automatic review). auto and again are the two switches, and a POST with either one, true or false, sets it. The answer also has max, each PR with its count, head, pending and waits, and news, what the relay said about the reviews it started, which the page shows once. - [Lead starts review](https://docs.building58.ai/api-reference/reviews/lead-starts-review.md): The Lead starts a review of a task's PR: {"pr": ...}. Only the Lead's token, under the rules in Tokens. - [List rounds](https://docs.building58.ai/api-reference/rounds/list-rounds.md): Past rounds, newest first, kept while the relay runs: id, game, time, spentUsd, seatNames and score only. A round that failed or was stopped has score: null. - [Start round](https://docs.building58.ai/api-reference/rounds/start-round.md): {"game", "seed", "settings", "seats", "watch", "capUsd"}: starts a round of a game, with a fresh session for each seat. 409 while a round runs, and 409 no_boundary where seats cannot get their boundary. 400 for a game that is not ok, with need when a need is what stops it, a seat count outside its p… - [Get seat steps](https://docs.building58.ai/api-reference/rounds/get-seat-steps.md): What one seat did, read now from Claude Code's transcript of its session: steps, each with kind (brief, prompt, says, runs with its tool, or result) and text, and the seat's tokens, usd and models. found: false before Claude Code writes it. For the round that runs only under Watch all (403 watch_non… - [Get current round](https://docs.building58.ai/api-reference/rounds/get-current-round.md): The round that runs, or null: its id, game, state, watch, spentUsd at API prices, unpriced (models with no price here), and its seats (seat, terminal, waiting: "trust" while that seat waits on the question about trusting its folder, or null, state, worked, tokens and usd). Never a brief, a role or t… - [Finish round](https://docs.building58.ai/api-reference/rounds/finish-round.md): Asks the game for its score and reveal, and ends the round. The answer has both. - [Stop round](https://docs.building58.ai/api-reference/rounds/stop-round.md): Ends the round with no score. - [Get settings](https://docs.building58.ai/api-reference/settings/get-settings.md): The settings on the page (see Limits and timings), or change them: {"values": {"wipLimit": 4}, "reset": ["leadCheckMinutes"]}. An unknown name gets 400 with unknown_setting, and a value out of range 400 with bad_value. Only the person. - [Set settings](https://docs.building58.ai/api-reference/settings/set-settings.md): The settings on the page (see Limits and timings), or change them: {"values": {"wipLimit": 4}, "reset": ["leadCheckMinutes"]}. An unknown name gets 400 with unknown_setting, and a value out of range 400 with bad_value. Only the person. - [List tasks](https://docs.building58.ai/api-reference/tasks/list-tasks.md): List or start tasks. - [Create task](https://docs.building58.ai/api-reference/tasks/create-task.md): List or start tasks. - [Get task](https://docs.building58.ai/api-reference/tasks/get-task.md): One task. - [Clean task](https://docs.building58.ai/api-reference/tasks/clean-task.md): Bring back its worker, or remove its checkout. If the checkout holds only files git ignores, Clean up answers 409 with error: "ignored" and the list in ignored. Send that list back as { "ignored": [...] } to delete them too. A file not on the list is asked about again. - [Move task](https://docs.building58.ai/api-reference/tasks/move-task.md): {"to": ...}: building, in_review, changes_requested, waiting_on_you or done. - [Resume task](https://docs.building58.ai/api-reference/tasks/resume-task.md): Bring back its worker, or remove its checkout. If the checkout holds only files git ignores, Clean up answers 409 with error: "ignored" and the list in ignored. Send that list back as { "ignored": [...] } to delete them too. A file not on the list is asked about again. - [List sessions](https://docs.building58.ai/api-reference/terminals/list-sessions.md): List or start sessions. A start can have a prompt, up to 8000 characters, which the relay types as the session's first prompt. A longer one is refused with bad_prompt. The list also has finished: the latest automatic reviewers that posted and left. Each session has its seat, the id of the person the… - [Start session](https://docs.building58.ai/api-reference/terminals/start-session.md): List or start sessions. A start can have a prompt, up to 8000 characters, which the relay types as the session's first prompt. A longer one is refused with bad_prompt. The list also has finished: the latest automatic reviewers that posted and left. Each session has its seat, the id of the person the… - [Type into session](https://docs.building58.ai/api-reference/terminals/type-into-session.md): Type into, resize or stop a session. A stop of the Lead is refused with 409 and lead unless the body says {"lead": true}, as the page sends once you confirm, so a script or test that stops sessions does not stop the Lead by accident. See When the Lead's session ends. - [Resize session](https://docs.building58.ai/api-reference/terminals/resize-session.md): Type into, resize or stop a session. A stop of the Lead is refused with 409 and lead unless the body says {"lead": true}, as the page sends once you confirm, so a script or test that stops sessions does not stop the Lead by accident. See When the Lead's session ends. - [Stop session](https://docs.building58.ai/api-reference/terminals/stop-session.md): Type into, resize or stop a session. A stop of the Lead is refused with 409 and lead unless the body says {"lead": true}, as the page sends once you confirm, so a script or test that stops sessions does not stop the Lead by accident. See When the Lead's session ends. - [Open session stream](https://docs.building58.ai/api-reference/terminals/open-session-stream.md): A session's output. - [Report context](https://docs.building58.ai/api-reference/terminals/report-context.md): Only with a session's own token, as its status line (relay/status-line.js) sends it: {"session_id", "percent", "size"}, how much of its context window that conversation has used. A session_id that is not the session's conversation now is refused with 409, and the relay key with 403. See How tired th… - [Forget failed session](https://docs.building58.ai/api-reference/terminals/forget-failed-session.md): {"sessionId"}: forget a session that could not come back after a restart. - [Set lead](https://docs.building58.ai/api-reference/terminals/set-lead.md): Make a plain session the Lead (terminal), which saves it and gives it the Lead's brief. Answer the offer to bring it back (asked: true; POST /terminals with resume and lead: true is what Bring back sends). With board: true or false, give or take away its permission to move the board. With starts and… - [List tickets](https://docs.building58.ai/api-reference/tickets/list-tickets.md): List or file tickets. The list comes in Board order: open tickets high, normal, low, in work order (rank) within each, then done and rejected ones, the latest changed first. It also has kinds (the kinds a ticket can have, see Kind), labels (the labels tickets have, A to Z, see Labels), wipLimit, lea… - [Create ticket](https://docs.building58.ai/api-reference/tickets/create-ticket.md): List or file tickets. The list comes in Board order: open tickets high, normal, low, in work order (rank) within each, then done and rejected ones, the latest changed first. It also has kinds (the kinds a ticket can have, see Kind), labels (the labels tickets have, A to Z, see Labels), wipLimit, lea… - [Get ticket](https://docs.building58.ai/api-reference/tickets/get-ticket.md): One ticket. - [Edit ticket](https://docs.building58.ai/api-reference/tickets/edit-ticket.md): Changes any of a ticket's fields, saved together, or none when one is refused: title (1 to 80 characters) and body (up to 8000) of a Triage or Ready ticket; priority, high, normal or low, which puts it last in that priority; kind, a word, or "" to take it away (Kind); labels, up to 5, at any status,… - [Move ticket](https://docs.building58.ai/api-reference/tickets/move-ticket.md): {"to": ...}: move a ticket that has no task, as Approve, Reject, Back to Triage and Merged elsewhere do (with pr for done). - [Order ticket](https://docs.building58.ai/api-reference/tickets/order-ticket.md): {"before": id} or {"after": id}: a Triage or Ready ticket goes next to another Triage or Ready ticket with the same priority. Usually only the moved ticket is written. When there is no room between two ranks, the Triage and Ready tickets of that priority are numbered again. If the store fails partwa… - [Decide proposal](https://docs.building58.ai/api-reference/tickets/decide-proposal.md): {"decision": "approve"} or {"decision": "decline"}: answers the ticket's proposal. Approving a replace, merge or close rejects the ticket and sets closedAs (replaced, merged or closed) and closedFor (the ids). Approving blockers sets blockedBy. 409 when no proposal waits, or when it no longer fits t… - [Propose ticket change](https://docs.building58.ai/api-reference/tickets/propose-ticket-change.md): {"action", "ids", "why"}: proposes to change a Triage or Ready ticket. action is replace (with one or more ids), merge (with one), close (with none) or blockers (the whole new list, which can be empty). why is up to 500 characters. Nothing changes but the ticket's proposal, which holds it back. The… - [Start ticket](https://docs.building58.ai/api-reference/tickets/start-ticket.md): Start a Ready ticket: a task, a checkout and a worker. A Lead's token is held to its limits. - [Approve ready tickets](https://docs.building58.ai/api-reference/tickets/approve-ready-tickets.md): Marks the Ready tickets that no person approved as approved by you, as Approve all Ready does: those in ids, or every one with no ids. It skips a ticket that is starting. Answers with the ids in approved. No session's token reaches it. - [Ask PM](https://docs.building58.ai/api-reference/tickets/ask-pm.md): {"idea", "repo"}: starts the PM. - [Get update](https://docs.building58.ai/api-reference/update/get-update.md): The relay's folder, and the update under way, if any, with who it waits for. rollback says which commit failed its check, why, and which commit runs, while the relay runs the commit it went back to, or while the folder is on it after a page-only step went back. While Update goes through merges one a… - [Start update](https://docs.building58.ai/api-reference/update/start-update.md): Fetches for the relay's folder and pulls when the plan says, as Update does. {"cancel": true} calls off a restart that is waiting, or stops a loop through merges between steps (One merge at a time). ## OpenAPI Specs - [openapi](https://raw.githubusercontent.com/bfalkowski/Building58/main/docs/openapi.json) This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.