Usage dashboard

The usage dashboard is the record of what your agents have spent: tokens, over time, broken down by model and by agent.

When to use it

Open it when the token badge on the Activity board is higher than you expected, before changing the model on a routine, or at the start of a week to see where last week went.

For what consumes usage and how the allowance works, read Plans and usage first. This page is about reading the numbers.

Getting there

1. Open the Usage tab on the home screen, or select the token badge in the Activity header.

2. Pick a period. 7 days, 30 days, or 3 months. The period changes the tiles and the charts; the yearly activity map always covers the current year.

3. Move between the four tabs — Overview, Models, Agents, Compute.

Overview

The tiles across the top answer “how much, how often, and with what”:

Tile	Reads
Total Tokens	Everything sent and received in the period
Requests	How many model calls that took
Active Days	Days in the period with any usage
Current Streak	Consecutive days of usage, with your longest as a hint
Top Model	The model that accounted for the most tokens
Usage Credits	This week’s spend in credits. The tile names the ratio it uses

Below them, Plan Allowance is the one to watch: a bar showing how much of the current allowance is gone, and a plain statement when it is exhausted. This is the same number the service checks before it starts a run, so when the bar is full, runs stop.

Daily Activity is a year-long map, one square per day, darker where you used more. It is the fastest way to spot a routine that started spending every day without you noticing.

Models and Agents

Both tabs have the same shape: a daily chart on top, a breakdown below.

• Models — which models the tokens went to. This is the biggest lever on cost, so this is where a decision to switch a model usually starts. See Choosing a model.
• Agents — which agents did the spending, and which models each one used. A routine-heavy agent shows up here long before you notice it anywhere else.

In both, the legend is a filter: turn a series off to read the rest without it, and Show All brings them back. Rows the chart does not have room for are collected into an “other” entry rather than dropped.

'Unassigned' is not a bug

The Agents tab has an Unassigned row for usage that is not attributable to one agent — account-level calls, including models you call directly through the LLM API. It is real spend with no agent to hang it on. The Sources tab splits it out.

Sources

The Sources tab splits the same tokens by where they came from:

• Agents — agent runs from web chats, routines, and channels.
• Agent API — agents called with an API key through the Agent API.
• endue Live — conversations visitors had with your agent on endue Live. Visitor conversations count as the agent owner’s usage.
• LLM API — models called directly with an API key through the LLM API.

Each row shows its tokens, share, and requests, with the models it used underneath. The legend works as a filter here too. LLM API calls are paid from purchased and promotional credits, while agent, Agent API, and endue Live runs draw on your plan allowance, so this is the tab to open when a credit balance moves and no agent explains it.

Compute

Browser sessions, workspace storage, media processing, and output storage are listed here but not yet measured. The tab says so on every row.

Read the empty rows as “not instrumented yet”, not as “zero”. Nothing on this tab is a claim that a resource is free.

One agent in detail

Expand a row on the Agents tab to find Open in Studio at the end. It opens that agent’s canvas on the Usage tab for the same period, where you can see chats and credits by source and by path (API key or channel connection). See Analytics.

Usage for one conversation

The dashboard is per account. For a single thread, type /usage in that conversation’s composer: a card shows input and output tokens, how many were cached, the weighted total, credits, and the models involved, with a link into the dashboard.

That card is a local view, not a message — it does not appear in the transcript, and it is gone after a reload.

Two different numbers, both correct

The /usage card counts what the conversation has spent so far. The context window panel counts what is loaded into the model right now. They are different axes and will not match.

Limits

• Usage is metered per account. The per-agent view is a breakdown of the same total, not a separate allowance.
• Figures settle shortly after a run, so the newest run can be missing for a short while.
• Runs made with your own key are not recorded here. A conversation that used only your own key reads as “no usage recorded” — which is accurate for Endue’s ledger and says nothing about what your provider billed you.
• Per-agent and per-source aggregation is not available in every environment. When it is not, the tab says so rather than showing zero.
• The Compute tab is a list of resources, not a measurement.

Related

Plans and usageWhat consumes the allowance, and how the weekly window works.

Choosing a modelThe decision the Models tab exists to inform.

Bring your own keyWhy some runs never appear here.

ActivityThe board whose token badge links here.