Usage & quotas
Track token usage and set daily budgets.
Two separate pages: Usage shows how much you've used, Quotas sets how much you're allowed to use.
Usage
Pick a time range (7, 30, or 90 days). The page shows:
- Summary cards: run count, total tokens, input tokens, output tokens — each compared against the same-length prior period.
- A daily chart.
- A breakdown table by agent, model, or run status — whichever dimension you want to look at.
A run is counted on the UTC day it ends, whether it succeeded, failed, or was cancelled.
The same report, narrowed to one agent and with today's budget for it on top, is also on that agent's own Usage tab (see Agents) — handy when you only care about one agent's traffic.
Quotas
Rate limits
Your organization has a ceiling on requests per minute — one for end-users, one for API keys — set by the Harnix team. Within that ceiling you allocate how much each is allowed; setting a value above the ceiling is refused.
Concurrent runs and queue depth
Concurrent runs shows how many chat runs are running now against your
organization's cap; Queue depth shows how many are waiting for a free
slot against the queue size (no queue means a chat starts only when a
slot is free). Only the Harnix team changes these; when the queue is full,
new chats are refused with queue_full (see Limits).
Background work such as document processing is not counted.
Daily token budget
An overall token ceiling for the whole organization, counted per UTC day, across all agents combined. Leave it empty for no limit. Each agent can also carry its own separate ceiling (see Agents) — both apply together, whichever is hit first blocks the run.
Once the budget is reached: a new run is rejected with budget_exceeded,
and a run already in progress is stopped before its next call to the
model.
This page also lists every agent's own budget in a table, editable inline.
Trace retention
The page shows Retention: how many days your organization keeps a trace's detailed content for, or Kept forever if the operator set it to never expire. This is set by the Harnix operator (30 days by default), not something you set here. Once it expires, only a trace's per-step content is removed — see Runs & traces for what a purged trace looks like.