Glossary

The vocabulary TokenTracker uses, in one place.

Fidelity — how a number was measured: exact (from the tool's own usage records), approx (derived), or estimated (e.g. a char-count fallback). Every figure is labelled so you know how much to trust it.

Identity — the four attribution levels: org → account → system (machine) → person (os user). The same person on two accounts and two machines still rolls up correctly.

Account — an AI provider login (e.g. an Anthropic or OpenAI account). A shared account used on many seats is surfaced for license planning.

Session — one continuous piece of AI work: a Claude Code session, a Codex thread, a chat conversation.

Turn — one step inside a session: your steering message or the model's response, with the tools and files it touched.

Trajectory — the build-story of a session: specify → build → correct → verify → accept, with corrections, pivots and verifications marked.

Working-style — your steering fingerprint: how much you specify vs. steer vs. correct vs. verify, plus signature phrases and traits. Stable across projects and re-derivable from the raw turns.

Decision — a consequential choice captured verbatim with provenance (which turn it came from), never invented.

Bridge — a file touched in two sessions; it links them in the knowledge graph — literally how work carries forward across time and tools.

The Brain — the knowledge layer (Layer 1): captured, distilled, searchable AI work, as opposed to Layer 0 telemetry (the numbers).

MCP — the Model Context Protocol; how your own assistant (Claude Desktop, Cursor, …) reads your tenant-scoped knowledge. See Use it from your assistant.

Ingest key — the per-org secret that enrolls a machine to your tenant. Stored only as a hash; mint and rotate freely in Admin.