4. AI: agents and local models
4.3 Agent limits and usage
How much of your plan you have used, when it resets, and the tokens and sessions of today and the last seven days.
In the app: «AI › Agents»
In short
- “Plan limits” shows how much of your plan you have used in each window and when it resets.
- “Usage” shows tokens and sessions for today and the last seven days, with a chart per day.
- The limits are read from the provider's servers every two minutes; “Refresh usage” reads them again right away.
Plan limits
For each limit window there is a box with the percentage used, a bar and “resets …”. The windows are “5 hours” and “Week” for Claude Code, “… hours” and “Week” for Codex, and per-model quotas for Antigravity. Above: the plan, the data source and when the data was read. Pi has no limit windows: the local models have none, and the kit does not query the cloud providers Pi may use.
| Color | When |
|---|---|
| Green | below 70% |
| Orange | from 70% |
| Red | from 90% |
The highest value also appears on the agent's row in the sidebar. For Claude Code, paid extra credits are shown separately (“Extra credits: … of …”).
| Message | What to do |
|---|---|
| not signed in: open … to see limits and usage | Open the agent in a project and sign in. |
| sign-in expired: open Claude Code to renew it | Open the agent: it renews the sign-in by itself. |
| API key: no plan limits | With an API key there are no limit windows. |
| Local models: no plan limits | Pi on the kit's local models: they run on your PC, with no plan. The usage below is still there. |
| The limits of Pi's providers are not read by the kit | Pi with a cloud provider: you see the limits on the provider's site or in Pi. |
| no data yet: the limits arrive with Codex's next reply | Use Codex once: the limits arrive with the response. |
| The server asks to wait: next reading … | The provider has limited the readings: the kit tries again by itself later. |
Usage
- “Tokens · Today” with the tokens of today and the last 7 days; hovering over it shows how many are new, how many come from the cache and how many are output.
- “Sessions · Today” with the sessions of today and the last 7 days.
- A bar chart for the 7 days, the last activity and the three most used models.
- Antigravity does not save tokens: for it only sessions are counted.
- For Pi the tokens are read from its sessions, including those on the local models.
Under the hood
- Sources of the limits: for Claude Code the Anthropic server, for Codex the OpenAI one (or, failing that, Codex's last response), for Antigravity the Google one. The kit uses the saved sign-in only to read: it does not renew it and does not send it anywhere else.
- Readings at most every 2 minutes, every 20 seconds with “Refresh usage”; after a “too many requests” response it waits as long as the server asks (from 2 to 30 minutes).
- Session files: data\home\.claude\projects, data\home\.codex\sessions, data\home\.gemini (Antigravity conversations), data\home\.pi\agent\sessions (Pi).