Private beta · built for teams running agent fleets on cron

Which agent is burning your Anthropic bill?

You've got a dozen agents running on cron — content, QA, research, retros. The invoice goes up every week and nobody can say which one did it, or whether last night's run was $0.04 or $4. Agent Cost Watchdog gives every agent run a real dollar figure, attributes it, and tells you the moment one blows its budget.

Get early access

The problem

Token counts lie. A run that shows 9 input tokens can still cost $0.04 because of cache writes — and cache-heavy agents look cheap right up until the bill arrives. Provider dashboards give you a account-wide total, days late, with zero attribution to the agent, the cron, or the run that spent it. By the time you notice spend is up, you're reading logs from a week ago trying to reconstruct what happened.

What it does

01 · METER

Every run, real dollars

Wraps your existing claude -p / agent CLI calls, reads the actual cost off the terminal usage event, and logs it per-agent, per-run — not estimated, not token-proxied.

02 · ATTRIBUTE

Per-agent, not account-wide

Every row is tagged to the agent that spent it. Answer "what did the CSO agent cost this week" in one query instead of an afternoon of log archaeology.

03 · BUDGET

Per-agent caps

Set a max-$-per-run ceiling per agent. A run that exceeds it is flagged the moment it finishes — not discovered on next month's statement.

04 · ALERT

Breach notification, not a dashboard you forget

Budget breaches push straight to you. No login, no report to remember to open — the alert finds you.

What the report looks like

AgentRunsTotalp90 / runStatus
content-agent14$3.82$0.41ok
retro1$2.31$2.31ok
curriculum-expert22$9.94$0.61ok
seo-agent6$6.05$1.87breach ×1

A real weekly rollup — every agent, every run, ranked by actual dollars spent, breaches called out. No dashboard to check; the breach found you first.

Who this is for

Anyone running more than a handful of scheduled Claude/GPT agents — content pipelines, QA crons, research loops, internal ops bots — who has looked at an Anthropic or OpenAI invoice and had no idea which agent to blame.

We built this to run our own fleet. It's the tool we needed before we could answer "why is spend up" without an afternoon of grepping logs.

Get early access

No spam, no mailing list. We'll email you directly when private beta seats open.