FAQ
Questions are written on this page only. The FAQ on the website's front page, the structured data for search engines and llms.txt are all generated from it; the questions marked for the site appear on the front page word for word.
The basics
Why do I need to apply?
Tokenman is in limited access for now: every account stands for a real machine and real quota, so we open it in batches, as many people as we can look after. To use it, apply: tell us what you want it for, and we usually answer within a working day — on your WeChat if you signed in with WeChat. An application keeps only how you signed in, what you wrote on the form and the decision.
How is it different from installing Claude Code myself?
It is the same official Claude Code; Tokenman sets up the machine, the skills and publishing, gives each task its own session, and keeps it running when you close your laptop.
Can I use it without coding?
Yes. Say what you want in plain words; changing code and opening pull requests is optional, and chatting, pages and documents work too.
Do I need my own Claude subscription?
Bringing your own is optional: use the subscription we provide, or connect your own Claude or ChatGPT subscription; either way you sign in with the subscription and need no API key.
What is the difference between hosted and self-hosted?
Hosted means we run a hub just for your company and you only connect subscriptions and invite colleagues; self-hosted means you install the same open-source software on your own machines, so the hub, the machines and the data stay with you.
Where do my code and files live? Are they safe?
They live on the machine that runs your sessions, in your own OS account that nobody else can read; subscription keys are kept by the hub and never enter a session, and actions are logged.
Does it work offline?
Sessions need to reach Claude's or ChatGPT's service to work; a self-hosted single computer needs no hub, and no GitHub either.
Is this an Anthropic product?
No. Tokenman is an independent open-source project (MIT) that runs the unmodified official Claude Code; Claude is a trademark of Anthropic.
Using it
How many sessions can one computer run?
There is no fixed limit by default: before a new session starts, Tokenman checks how much memory is left and how busy the machine is, and a session that does not fit waits; running sessions are never touched. For a hard limit, set FLEET_GLOBAL_MAX_SESSIONS in the configuration.
If I close the terminal, do my sessions keep running?
Yes. Sessions run on the machine, not in your terminal; type fleet again and everything is where you left it — you land on the session you looked at last (from any computer or phone), or on a new one if that session has been closed. fleet quit closes only the client.
Can I use it without GitHub?
Yes. Sessions that only chat or make pages and documents do not need GitHub; to let a session change code in a repository and open pull requests, install gh and run gh auth login. GitHub only hosts the code and the pull requests — tasks and sessions live on the hub; what carries on and what waits when GitHub is unreachable is in Troubleshooting.
A session stopped on a question. What now?
It is most likely waiting for your answer (pick an option, approve an action). The ⌘P panel lists those sessions under "Waiting on you"; select one to see its question.
Self-hosting
What is the least I need to self-host?
One macOS or Linux computer, a Claude or ChatGPT subscription, and the claude or codex command line signed in once. One computer needs no hub; the self-host quick start gets you to a first session in about ten minutes.
When do I need my own hub?
When you want to connect several machines, invite colleagues, or watch sessions from the web console or a phone. The hub is one program on one port that stores its data in SQLite or Postgres; see Install and deploy.
Does a self-hosted hub show the website?
No. A self-hosted hub opens straight into the console; the website runs only on our own hub.
How do I upgrade? Does it interrupt running sessions?
Machines and clients follow the stable release: fleet update catches up at once, and they catch up by themselves anyway. A switch replaces the programs and background services, never a running session.
How do I find out what is broken?
Run fleet doctor first; what each line means and what to do when it is red is in Troubleshooting. To have someone else look, send the output of fleet doctor --json to your admin.