Why it matters
AI spend is easy to lose sight of when several agents and people use it. You need to know what was called, who funded it, and how much allowance is left before it runs out.
What you can do
- See credits used and credits remaining in the current period, and when the pool resets.
- See which calls ran on LakeLogic's key and which ran on your own key or a self-hosted model.
- Break usage down by agent, model and day.
- See what your own AI assistant did through MCP, which uses no credits.
How to do it in LakeLogic
- Open AI → Usage & Cost (
/ai-usage). Organisation admins can see it. - Pick the window with the date control: a preset or a custom range. The default is your current credit period, the same window the Billing page uses.
- Narrow to one kind of call with the type filter (All Types by default).
- Use Show above the charts to pick the measure the charts use.
- To cut spend, move Zeus to your own provider key or a self-hosted model under AI → Models & routing. Those calls use no credits.
What you get
Summary tiles
| Tile | What it shows |
|---|---|
| Model calls | Individual AI requests in the window |
| Tokens | Total, split into prompt and completion |
| Managed AI usage | The value of credits used, or how many calls could not be priced |
| Zeus credits used | Credits used by calls on LakeLogic's key |
| Credits remaining | What is left of this period's pool and when it resets. Enterprise shows Agreed. |
| Average latency | Request to response. not measured when there were no calls. |
| Active threads | Ongoing Zeus conversation threads |
There is no estimated cost or failed request tile: neither is measured here, so neither is shown as zero.
Charts: AI usage over time (optionally By agent), Usage by agent, Model breakdown and Response time by agent (median and p95). A day still in progress is shown faded.
Model calls table: query, type, agent, provider and model, tokens, latency, rating and time, plus:
| Column | Meaning |
|---|---|
| Funding | LakeLogic-managed, Customer API key or Customer-hosted |
| Credits | Credits used. 0 on your own key. |
| Usage (£) | Managed calls only |
| Provider cost | The provider's price for the tokens. On your own key it is an estimate; your provider bills you. |
Assistant (MCP) calls: calls, refused calls and proposals (approved and rejected), and a table per API key with calls, refused, tools used and last call. These use no Zeus credits and show no cost. See Connect your AI assistant.
Good to know
How credits are counted:
- Each call is priced from its tokens, at the model's price in force when it ran.
- Only calls on LakeLogic's key use credits.
- Which key a call used is recorded when it runs. Older calls are inferred from your current AI settings.
- Credits are worked out per call, to two decimal places, and summed.
- A call whose model has no price is counted as could not be priced, not as 0.
The Billing page reads the same function, so both show the same figures. Allowances per plan are in Get answers about your estate in plain English.