SemAgentUsage metrics
Admin Panel → Account → Usage shows your consumption. The top bar shows your plan's name (Plan: …); next to the title are the period picker (This month / Last 30 days / Custom range) and the Refresh button; the Table / Chart buttons under the metric cards switch the view. This page explains what each metric means.
Credit balance
When the credit model is active you see a single Credit balance card at the top: N credits left, Used this month: used / allowance (≈ $) and a Lasts about N days at this rate estimate (with the last 7-day daily average under it). If you bought extra credits, what's left of them shows as Add-on package. Without the credit model, a Cost card for the period takes its place, and the Tokens, Reasoning calls and Q and A generation calls cards show as well.
1 credit = $0.01. All your usage — chat, voice, knowledge-base processing — is drawn from one credit balance; crawling pages costs no credits. So instead of "which row costs what", there's a single, understandable unit.
A failed call is not charged — an operation whose answer never reached you costs no credits. The top-up you buy is used after your monthly balance runs out.
The What spends credits? expander lists each operation's credit price. Fixed-price ones show 1 credit or N credits, metered ones pay-as-you-go (max N). The main credit-consuming operations:
- Chat/assistant: chat message, reasoning, Q&A generation, smart content processing.
- WhatsApp and Instagram channels: the AI's answer to a customer's message costs the same as a chat message in the widget. The “Talk to a person” button, and on WhatsApp the short reply sent for a photo, voice note or file with no text, cost nothing (the AI is not called). Each stored incoming file costs 1 credit (Channel file (WhatsApp / Instagram) in the list). Meta's WhatsApp message charges don't use credits; they are billed to your WhatsApp account (Messaging › Charges).
- Answering from the knowledge base: the query expansion and reranking behind an answer are not rounded up call by call; their real cost over the month is bundled into credits.
- Knowledge base: KB embedding (per page), Q&A indexing (per page), storage (GB/month). Processed pages aren't charged one by one: the month's embedding pages are added up and written as 1 credit per 25 pages (Embedding bundle (25 pages) in the list). Crawling pages costs no credits.
- Voice: voice reply (TTS), speech recognition (STT).
- Memory: user memory, conversation summary, memory service, personal briefing, skill learning, quality patterns, security scan.
Resource metrics (label → meaning)
Resource metrics tracked separately from credits. Every card with a limit shows Remaining / Total and a bar; the same numbers appear in the Limit summary table with when each was last used:
| Metric | What it measures |
|---|---|
| Messages | Total messages in the period (chat and widget counted together). |
| Bandwidth | Bandwidth consumed for web pages fetched through the proxy. |
| Crawled pages | Total pages indexed via sitemap (per month). |
| Embedding pages | Embedding page usage of ingested/searched data (1 page ≈ 750 tokens); counted against the plan embedding limit. |
| Storage | Total disk space used by the knowledge base + embeddings. |
The message counter is not a limit. No per-message limit applies; usage is billed through credits. That's why the Messages card reads Unlimited — billed through credits and the Limit summary table shows Unlimited.
Hard ceiling & no surprise bills
Limits are a hard ceiling: when reached, new requests stop; no automatic overage charge. The widget shows your customer a graceful redirect, not an error; it resumes the moment you upgrade. To buy more: Subscription & Billing → Buy Extra Credits.
Conceptual background (why the credit model): Credits & usage. Plan limits and plan-gating: Plans.