Docs / Usage and limits

Usage and limits

ThreadGrid runs on your own Claude Code, so the work counts against your Claude plan or your API key, the same as using Claude Code yourself. There's no ThreadGrid bill. This page explains what you can see and the limits you can set.

What you can see

Settings → Usage & limits shows:

  • The token meter. Tokens used today, with a bar against your daily limit when limits are on.
  • The last 14 days. Tokens per day as a chart or a table.
  • Where today's tokens go. Split by the kind of work: conversation, planning, building, review and so on.
  • The last 30 days. Tokens, estimated cost, runs, tasks shipped and the costliest tasks.
  • Per task. Every task page shows its own tokens and estimated cost.

The inspector (⌘I) shows today's tokens with a 14-day trend as well.

Token counts add input and output tokens; cached reads count at a tenth, since that's roughly what they cost.

Limits are off by default

Out of the box nothing is capped by dollars or tokens. The team works until your Claude plan says stop, then waits and picks up where it left off (see usage limits). Each run still has a turn cap, so no single run can go on forever.

If you'd rather have a hard ceiling, turn limits on.

Setting limits

  1. Open Settings → Usage & limits.
  2. Turn on Set limits.
  3. Set any of the limits. Leave a field empty for no limit. Amounts accept shorthand such as 500k or 2m.
LimitWhat happens when it's reached
Tokens per dayCounted across the project. No new runs start until midnight, local time; running steps finish.
Spend per dayThe same, measured as estimated cost.
Tokens per taskThe task pauses at the next step and asks you: Raise limit or Stop here.

While a limit holds things up, waiting work is marked Budget. Nothing is lost: raise the limit or turn limits off and it picks up where it stopped. If you set cost limits on goals in 2.x, they're kept and apply again once limits are on.

Spending less

ThreadGrid 3.0 already does a lot to keep usage down:

  • The Lead triages asks in the conversation, so there's no separate triage run.
  • Light work such as conversation, triage, release notes, small reviews and QA of small tasks uses a faster, cheaper model, unless you've picked a model for that teammate.
  • Small tasks get one combined review and QA pass instead of two.
  • Fix runs pick up the build session where it left off instead of starting again.

You can do more:

  • Pick the Fast profile in the composer for routine asks.
  • Set each teammate's model and effort in Settings → Models & effort.
  • Lower concurrency in Settings → Autonomy & gates (3 tasks at a time per project and 4 across the app by default).

Several Claude or Codex accounts

Add more than one account in Settings → Accounts and ThreadGrid moves work to the next one when an account reaches 85% of its limit. Each account shows In use, Standby, Resting or Limit reached. If every Claude account is resting, you can choose to keep going on the other agent (Codex), and the other way round. Sign-ins stay in the macOS Keychain.

Something unclear or out of date? Email [email protected].