For AI product teams
Ship LLM featuresknowing what each call costs.
Every model call shows up inside the request that made it, with its tokens, cost and latency. When agents dig through your telemetry, FerrisKey decides what they can touch.
Did that prompt change help?
Pick a change and compare cost, tokens, latency and user feedback, measured on real requests before and after.
prompt v12→prompt v13· last 24 h, 18k requests each
| metric | prompt v12 vs prompt v13 | change |
|---|---|---|
| cost / 1k requests | $4.10$6.90 | 68 % |
| input tokens | 2,4004,300 | 79 % |
| p95 latency | 2.1 s2.3 s | 10 % |
| thumbs up | 81 %82 % | 1 % |
v13 costs 68 % more and buys one point of satisfaction. The extra 1,900 tokens come from the new examples block, so trim that before rolling out.
Where the time goes today.
The steps it takes now, next to the steps it takes with FerrisTrace.
Today
3 steps · 3 tools- 1provider consolesee the total
- 2spreadsheetmap API keys to features
- 3githunt for the prompt change
With FerrisTrace
1 step · 1 tool- 1FerrisTracespend broken down by feature, so the spike points at its cause
You see cost per feature as it moves, before the invoice tells you.
Start with these.
- AI observabilityWhat each LLM call cost, and how long it took.Learn more
- AI agentsAgents investigate with scoped FerrisKey tokens.Learn more
- Traces, logs & metricsFollow one request from the click to the database.Learn more
- Incident responseFrom the first alert to the postmortem, on one timeline.Learn more
Find out what your AI features cost to run.
Send your LLM calls over OpenTelemetry and their cost and latency show up next to the rest of your traces.