LLM Analyser
Optimize → LLM Analyser
Where NudgeBee's own AI spend goes. Every investigation, agent run and tool call has a token cost; this is the breakdown, by conversation, model, agent, tool and user.
The tab is gated by the per-tenant LLM_ANALYSER feature flag — see Tenant Settings → Features. If nobody can see it, that flag is the first thing to check.
Views
| View | Answers |
|---|---|
| Overview | What is total spend, and how is it trending? |
| Conversations | Which investigations cost the most? Drill into any one for its per-step cost. |
| Models | What is each model costing, and how much traffic does it take? |
| Agents | Which agents are expensive? |
| Tools | Which tool calls are driving token use? |
| Users | Who is generating the spend? |
| Cost Report | A scheduled summary, delivered on a cadence you set. |
Group by day, week or month. Filters combine with Any of or All of, so you can isolate, say, every conversation that used a particular model and a particular agent.
Conversation detail
Opening a conversation shows the cost of each step, which is how you find out that one expensive investigation was a single agent looping on a tool rather than the model being pricey.
Cost reports
Schedule a recurring account cost report so the numbers arrive without anyone opening the tab. Configure the schedule from the Cost Report view.
Model pricing
Spend is computed from a model price list. Where you use a model NudgeBee does not have a price for — a self-hosted model, a negotiated rate, a new release — set the price yourself so the totals stay real. Prices are per tenant.
LLM Analyser or AI Gateway?
They measure different things and both can be true at once:
- LLM Analyser — what NudgeBee's own agents spent, broken down by the platform concepts that caused it (investigation, agent, tool).
- AI Gateway — what your organisation's tools spent going through the gateway, broken down by caller.