Company-wide usage analytics for AI coding agents — Claude Code, Codex and OpenCode. See who is using them, on which projects, with which models, and what it would cost, across your whole team.
This is the server. It runs at hototel.com (formerly hotusage.ai, which still works), so most people never install it — you just need an invite.
Your admin sends you an invite link. Open it, pick a password, then install the client on each machine:
curl -fsSL https://raw.githubusercontent.com/hotdata-dev/hototel-client/main/install.sh | shYour browser opens, you approve the machine, and it starts reporting. From then
on it syncs by itself. Headless machines run hototel signin instead, which
prints a URL and a code to approve from any browser.
Two ways to see the numbers: open the dashboard, or ask your coding agent — the installer teaches Claude Code and Codex to answer questions like "what did we spend on Claude Code last month?"
Filter by person, project, tool and date range. The daily chart stacks either by token type or by person, in tokens or dollars, with a table view of the same numbers.
The dollar figures are API list-price equivalents, not a bill. They are
worked out from token counts at published rates. If your team is on a Max, Team
or Enterprise plan you pay a flat per-seat fee and none of this — an
organization can easily show tens of thousands here while paying a few hundred.
Use them to compare people, projects and trends; for actual spend see the spend
report at claude.ai/admin-settings/usage, which exports per-user cost as a
daily CSV.
The Insights page, linked from the header, is the dashboard you can talk to. Ask a question in plain words — who worked on the billing service this month?, which sessions burned context without producing anything? — and a model answers it by running SQL against your organization's usage, listing every query it ran under the answer so a number can be checked rather than trusted.
Above the chat, Takeaways is the same analyst's read of the last 30 days, sorted into three columns: good work (practices worth spreading), valuable (where the effort concentrates and appears to pay off) and redundant (duplication or waste worth a look). It regenerates on demand and otherwise every six hours.
The model only ever sees your organization's database, and only members signed in to the dashboard can ask. On hototel.com the model is an open-weight one hototel runs itself, so your usage is not sent to a third-party AI provider. The page is off until the operator connects a model, and says so when it is.
Everything is on the Organization page, linked from the header.
Inviting people. Two kinds of link, which you share yourself — nothing is emailed:
- a single-use invite bound to one address, valid 7 days
- a team link anyone can use to join with their own email and password
You can restrict a team link to an email domain and cap how many times it is used; it expires after 30 days by default, 90 at most. Restrict it to your domain unless you are sharing it privately — anyone holding an open link can join and read your organization's usage. Every new account has to confirm its email address before it can see or report anything, so a restricted link really does admit only people who can read mail at your domain.
Managing members. Promote or demote admins, and remove people. Removing someone revokes their logins and their machines; usage they already reported stays.
Revoking access. Outstanding invites and signed-in machines are both listed with revoke buttons. Each machine shows what it may do and when it was last active, which is how you spot one that has quietly stopped reporting.
Someone can belong to several organizations; the account menu switches which one is active — the org their dashboard shows and their machines report into.
- Usage from an address that isn't a member is rejected. The person's client shows the error; once you invite them, their next sync succeeds.
- Cursor and Gemini CLI are not covered. Neither keeps local token counts, so there is nothing to read.
- First load after a quiet spell can take 10–20 seconds. An idle organization's database scales to zero and has to wake up. Normal, not a hang.
- Each organization's usage is physically separate — its own database, not a filtered view of a shared one.
Running your own instance, or working on hototel itself?
See docs/operating.md.