What is TokenTracker

TokenTracker measures and remembers your team's AI work across every tool — Claude Code, Codex, Cursor, the chat apps, the APIs — and turns it into something you can actually use.

It has two layers:

Layer 0 — Telemetry

Where the tokens go. Every machine runs a small agent that ships a daily, numbers-only rollup: how many tokens, on which tool, model, project, account, person and machine, with an estimated list-price cost. No prompts, no code, no message text — just the measured usage. This is the fleet view: budget, cost control, license planning.

Layer 1 — the Brain (knowledge)

What the work actually was. When an org opts in, the agent also captures your AI sessions — redacted transcripts distilled into the build-story (specify → build → correct → verify), your working-style fingerprint, and the decisions you made — all linked into a knowledge graph and searchable (including from your own assistant via MCP). This is the moat: the reusable memory of how your team works with AI.

Telemetry answers "how much, where, who?". The Brain answers "what did we figure out, and how do we reuse it?".

Start with Install the agent, then read Concepts.