Documentation
Put your first agent to work tonight, then learn every dashboard feature step by step: projects, launches, rotation, previews and Vault.
Get started
- Get startedSign up, get your Compute and start an agent in your own repository. Shut the laptop for the night; the agent keeps working until you're back.
- Projects and launchesEvery repository becomes a project on your Compute at $0. Launch Claude Code, Codex or OpenCode in it with one click, each on its own subscription or API key.
- Connect Claude and Codex subscriptionsConnect your ChatGPT and Claude subscriptions once. Agents start signed in, you switch a running one from its terminal, and every limit shows its reset.
- Add API keysRun your agents on your own key from Anthropic, OpenAI, xAI, DeepSeek, Z.ai, abliteration.ai or Gemini: checked on save, renamed, replaced and watched for balance.
- Choose and set up your agentClaude Code, Codex and OpenCode ready in every terminal, with approvals off, your instructions read and the tools agents reach for already installed.
Recipes
- Run an overnight refactor and wake up to itHand an agent a long refactor before bed, close the laptop, and read the finished branch, its report and the night's log over coffee.
- Stretch several subscriptions through a nightPut two or more Claude or Codex subscriptions in one rotation, continue at each reset and keep a reserve for your own day, so the agent runs till morning.
- Answer your agent from the lock screenThe agent's question reaches your phone with its choices as buttons. Answer 1, 2, 3, yes or no from Telegram or a push, without opening the dashboard.
- Drive your agents by voice from ChatGPTConnect detachbox to ChatGPT once, set a dot, its always-on agent, to watch your terminals, then on the road ask who waits, answer by voice and hand out the next task.
- Show a client your dev serverTurn the dev server your agent runs into a public HTTPS link or a QR code for a client, then close it again with one switch.
- Hand the agent a screenshot or a phone photoShow the agent the broken layout, the error dialog or a whiteboard sketch: paste a screenshot, drop a file, or send a photo straight from your phone.
- Run several agents at once on separate projectsBring each repository in as a project, give each its own agent and subscription, start them side by side and move between them with one key.
Work with your agents
- The browser terminalTerminals that keep running after you close the tab: open, rename, restart and end them, copy text out, search the scrollback and open the pages your agent opens.
- The terminals listSee at a glance which agent is done, which needs you and which is still working, in the order you set, with filters, search and number keys.
- Keyboard shortcuts and the command menuReach any terminal, project, account or action from the ⌘K menu, jump between agents with number keys and use the terminal's own keys on a Mac or a PC.
- Files, screenshots and photosPaste a screenshot and the agent sees the bug. Drop files on the terminal, send phone photos, open the files the agent prints and share one by link.
- Preview URLsEvery dev server on your Compute gets its own HTTPS address: open it in your browser, scan it onto your phone, make the port public for a client.
- Drive your agents from a phoneAnswer your agents from the phone: one-tap replies on the key bar, dictation, photos straight from the camera and an app icon that counts who waits.
- NotificationsA push, Telegram message or email when an agent asks, finishes or stops on a limit. Answer its menu from the lock screen; hold pings at night.
- Drive your agents from ChatGPT, Claude, Grok and MuseTalk to your agents from the car. A dot, ChatGPT's always-on agent, watches every terminal and writes when an agent asks; you answer by voice and hand out the next task.
Keep it running
- Continue after the limit resetsLet an agent stopped by a subscription limit carry on by itself: detachbox types "continue" into its terminal the minute the limit resets.
- Rotate subscriptions and keysWhen one Claude or Codex subscription hits its limit, the agent moves to the next and keeps working. Your order or soonest reset first, at a line you set.
- Keep a reserve for yourselfAgents spend your subscription and stop at the share you keep for your own work. Lift the reserve before the weekly reset and they use up the rest.
- CPU, memory and diskWatch your Compute's CPU, memory and disk from the top bar, find the terminal that uses the most, clear package caches and get past a full disk or a memory stop.
- Give agents your service keys with VaultLet agents call Stripe, GitHub, Cloudflare and other services through Sallyport, without the keys ever reaching your Compute. Each project gets its own Sallyport agent.
- Activity logSee what happened overnight: every rotation move, continue after a limit, terminal and sign-in on your account, with the exact time and who did it.
Compute and billing
- Your ComputeOne Compute runs all your agents: a fresh Linux environment with Docker in every terminal, one CPU and memory budget for all of them and a disk your projects live on.
- Account and sign-inSign in with Google, a password or a passkey, add two-factor sign-in, see your devices, reset a password, and pick the dashboard's language and theme.
- Plans, billing and supportOne monthly price from $39, every project included. Compare sizes up to Business Ultra and Enterprise, change size or cancel, earn credit, reach support.
- Troubleshooting and FAQThe fix first for common detachbox snags: previews with nothing listening, signed-out accounts, limits, memory, pings, pastes, a full disk, an unpaid Compute.
Enterprise on your server
- Enterprise on your own serverYour code stored and built on your own Ubuntu server, the same dashboard for your engineers, the node kept current by us. What runs where and how to get one.
- What the server needsThe server, disk and network a detachbox node needs: Ubuntu Server 26.04, 4 cores, 17 GiB of RAM, an empty second disk and outbound access only.
- Install the nodeOne command turns a clean Ubuntu server into your node in 30 to 60 minutes: its checks, its eight steps, a rerun after a stop, the log and what lands where.
- What terminals reachTerminals on your node reach the internet and only the private networks your admin opens on the server itself, never the node; plus the node's addresses.
- Keep the node runningCheck the node with doctor, let it update itself, choose when Incus and cloudflared move, handle kernel updates and reboots, and give the server more power.
- Remove the nodeTake the node off your server with one command, keep the pool with your account's files or destroy it too, go on after a failed step, or do it all by hand.
- Node troubleshootingThe fixes for a failed check or install step on your server, the logs that say why, and what to send detachbox support when you need a hand.