“Why did the AI bill double?”

1. 1

   Install

   Sign in with Google, then run one command on each engineer's machine. The collector is open source and short enough to read.

2. 2

   Collect

   It reads the logs Claude Code and Codex already write and pushes a per-day summary every hour. Vendor-reported estimates sync from your Claude Console and are never summed with the local estimate.

3. 3

   Answer

   Login imports available history immediately when automatic collection is on, then pushes hourly. Otherwise run kibble push. Owners see totals per person, team, model and repo; each engineer sees their own machines. Visibility, never surveillance.

From the example week. Made-up numbers, real arithmetic.

The short answers, in one place. The long ones are in the collector's source.

What does Kibble show?

What every coding agent on your team spent this week: spend per engineer, team, model and repository, this week against last, with the biggest mover named. It covers every laptop, CI runner and cloud box that runs the collector, plus vendor-reported estimates read from your Claude Console. A Capabilities page shows observed Claude Code skill usage, MCP calls and estimated description context. Codex records explicit skill selections and recognized command submissions; automatic skill reads and skill cost attribution remain unobserved. Zero does not mean unused. Kibble is a usage dashboard for software teams, not a bank, wallet, card or payments product.

Which coding agents does Kibble track?

Claude Code and Codex from their local usage logs today, plus Claude Code usage analytics from your organization's Claude Console. Vendor analytics supplies usage and estimated cost for supported records, not an invoice.

Does Kibble read prompts or code?

No. The collector sends counts only: token totals, model names, opaque session ids and repository names. Never prompts, file contents, tool arguments, paths, hostnames or hardware ids. The server's ingest schema is strict, so a field it does not expect is a rejected request. The collector is open source and short enough to read.

How is Kibble installed?

Sign in with Google at app.usekibble.com, then on each engineer's machine run npm install -g @usekibble/cli && kibble login. The CLI derives daily summaries from the session logs the agents already write. When automatic collection is enabled, login sends available usage immediately and schedules hourly pushes. The first sync includes today and the preceding 29 UTC days where local logs exist. If automatic collection is off, run kibble push manually.

How much does Kibble cost?

Free covers one person, two machines, every agent, and the last 30 days of history. Pro is $100 a month or $1,000 a year for up to 30 people, then $5 per person a month past that, with all history, teams, roles, invitations and vendor-reported estimates from your Console, before tax, and the first subscription includes 30 days free with a card required. Self-hosting is on the roadmap: every feature on, nothing metered, on your own infrastructure.

Is the local spend number the same as the invoice?

No. Local rows are estimates priced at list rates, and Claude Console analytics also reports estimated cost. Kibble uses vendor estimates for matching daily usage and never adds overlapping estimates together. Check your vendor invoice for actual charges.

Can engineers see each other's usage?

Each engineer sees their own machines. Owners see totals per person, team, model and repo. Managers see the teams an owner assigned them. Devices are visible to their owner only.

Can Kibble be self-hosted?

Not yet. Self-hosting is on the roadmap: Kibble on your own infrastructure, every feature on, nothing metered. The CLI collector is open source today, so what leaves each machine can already be read.

How this differs from a proxy or a per-laptop tool is on the comparison page.