Usage and Quota
Use Usage & Quota to see how much of your platform allowance you have used, which agents are consuming it, and which requests contributed to the total.
Open the ☰ drawer and select Usage & Quota.
The page has three main cards:
- 24-Hour Rolling Quota — your current quota status.
- Usage by Agent (Last 24h) — a visual breakdown by agent.
- Recent Usage History — a sortable, paginated request history table.
The page header also includes a Refresh button. Click it to reload the latest usage data without refreshing the browser tab. The numbers shown reflect the last page load or refresh; they are not a live stream.
24-Hour Rolling Quota
The Usage & Quota page leads with a quota card showing:
- A progress bar showing usage versus your quota limit.
- A fraction such as
1,111 / 200,000 units. - A percentage label such as
0% used. - Used, Remaining, and Limit stat tiles.
- A status tag.
| Tag | Bar color | Meaning |
|---|---|---|
| Good | Green | Under 70% used. |
| Warning | Yellow | 70–90% used. Heads-up. |
| Critical | Red | 90–100%. You're almost out. |
| Quota Exceeded | Red | Over 100%. Chat input is disabled until the rolling window resets. |
The quota uses a 24-hour rolling window. Usage does not reset at midnight. Instead, each request drops out of the calculation 24 hours after it happened.
This means headroom returns gradually as old activity ages out. If usage spikes at 10:00 AM, that spike stops counting around 10:00 AM the next day.
Usage by Agent (Last 24h)
This card shows each used agent as a labelled horizontal bar with a unit count. It is a visual breakdown, not a table.
Use it to identify:
- Agents that receive the most traffic.
- Agents that consume more quota than expected.
- Agents that may be using more expensive models or longer responses.
Agent names may appear as internal identifiers, such as esp_rainmaker_control, rather than display names.
Recent Usage History
The Recent Usage History card shows individual usage records in a sortable table. Click a column header to sort by that field.
| Column | What it shows |
|---|---|
| Time | When the request happened. |
| Model | The AI model used for the request. |
| Type | The request type, such as chat. |
| Input | Raw input token count. |
| Output | Raw output token count. |
| Quota | Normalized quota units charged for the request. |
| Status | Request outcome, shown as a status code badge. |
The table includes pagination controls at the bottom, including rows per page, the current record range, and previous/next page buttons.
Use this table to investigate unexpected usage. Compare Input, Output, Model, and Quota to see whether high consumption came from long prompts, long responses, model choice, or repeated requests.
The Status column uses HTTP-style status codes. A green 200 means the request succeeded. Non-200 values indicate that the request did not complete normally and should be investigated with your platform admin or logs.
What counts as usage
Quota works like a spending budget, not a token counter. Every request draws from your balance based on what it actually cost to run, which depends on the model, modality, and how much of the context was already cached. Output tokens are generally more expensive than input tokens, so longer AI responses cost proportionally more. For example, a request with 500 input tokens and 100 output tokens has 600 raw tokens, but may charge more quota units if it uses a higher-cost model, the quota reflects the real cost of running that model, not the raw count.
Usage can include:
- Text input and output.
- Cached context — reusing context from a recent turn costs less than fresh input, which lowers quota spend for longer ongoing conversations.
- Voice or audio processing.
- Tool calls.
- Knowledge Base retrievals.
- Attachment analysis, such as images or documents.
You can see exactly what was charged per request in the Quota column of the Recent Usage History table. The Input and Output columns show raw token counts for reference — the Quota column is what actually comes off your balance.
If you need the exact normalization rules for your deployment, ask your platform's admin. In self-hosted deployments, the configured models and quota policy determine how units are calculated.
When you hit the quota
When you reach 100%:
- The chat input is disabled in any open chat.
- The page shows Quota Exceeded.
- The agent's reply usually says something like "Daily quota exceeded" if you try to push through.
- A contact email may appear so you can request a quota raise.
What to do
- Wait it out. The quota is a 24-hour rolling window. As old usage falls off, you regain headroom — usually within a few hours.
- Email Espressif at private-agents@espressif.com if you need a permanent increase or hit the cap regularly. Mention your account email and what you're using the platform for.
- If you self-host the platform in your own AWS account, ask your platform admin to raise your cap directly — the contact email is for Espressif's hosted platform users.
Privacy
- Your usage is visible to you and to platform admins.
- Other end users cannot see your numbers.
Usage data may include model names, timestamps, request types, and agent identifiers.
Tips
- Watch for trends. If you cross Warning repeatedly mid-day, you'll likely hit the cap before the day ends.
- Check the history table. It shows whether usage came from input size, output size, model choice, or repeated calls.
- Voice can be heavier than text. Long voice sessions may consume more normalized units than short text chats.
- Big attachments cost more. Vision and document analysis can use many units. Crop and trim before sending.
- Switch to a cheaper agent for casual questions. If a quick reply is enough, find an agent on a smaller model.