AI observability

Know what every LLM callcosts, and why.

Each model call is recorded with its tokens, cost and latency, broken down by feature and customer, and tied to the request that made it.

LLM spend · last 7 days · $755

Doubled on Tuesday: a prompt change added 6k tokens of examples to every call.

Change the grouping and click a bar to see what is behind that number.

Simple questions, slow answers.

What it takes today to answer basic questions about your LLM usage.

Today

4 steps · 4 tools

With FerrisTrace

2 steps · 1 tool

You can explain the spike before the invoice lands.

Cost per callWorked out from prompt and completion tokens, at each model's rate.
Spend by any dimensionGroup by feature, model, customer or environment.
Latency in contextSee model time inside the request, next to retrieval and your own code.
Spend alertsGet an alert when a feature or a customer goes over the threshold you set.
Any providerBuilt on the OpenTelemetry GenAI conventions, so you can change model vendors.
Find any callLook a call up by user, trace id, model or prompt version.

Send your LLM calls over OpenTelemetry and the cost and latency come with them.