ClawIndex

How much are you spending on stuck agents — and agents escalating when they didn't need to?

Open-core observability for AI agent fleets. Built for the person who owns the bill, not the person who owns the pager.

The problem

Agents are running, something is wrong, and there is no visibility into which agent or what it is costing. SRE dashboards show what is slow. ClawIndex shows what is expensive, what is failing, and what cannot be seen at all — spans and agents with no registered identity that pass through your infrastructure without a trace.

What it does

Fleet economics

Real cost estimates derived from actual token usage, not guesses. Every agent, every span, attributed to the operator who ran it.
Fleet-relative comparisons

Which agent is running at 4× the typical cost for its kind. No built-in alerts, no severity scores — just the number and the fleet baseline it came from.
Unattributed telemetry

Agents and spans with no registered identity, surfaced as a governance finding. Dark activity you are paying for and cannot currently account for.

Design principle

ClawIndex does not editorialize. There are no severity scores, no verdict labels, no thresholds baked in. The dashboard reports numbers and fleet-relative comparisons. You supply the business context; the tool supplies the facts. Whether a cost is acceptable is your call, not ours.

Open core

The collector and dashboard are open source under AGPL-3.0 and self-hostable. No account required to run your own instance.