4 min read

Set up AI usage metering

Enable AI metering for your organisation, get your people connected with one settings change, and know what to expect when data starts flowing.

Updated 12 July 2026

EngLedger meters AI usage at the request level: every request records its cost, tokens, model, provider, latency and the person behind it, whether it went to Anthropic (Claude), OpenAI (GPT) or any endpoint that speaks the OpenAI-compatible API, including Gemini and self-hosted models. Setup is a settings change, not an implementation project: each person points their AI tools at a personal access URL and keeps working exactly as before, with their own API keys.

Enable metering for your organisation

  1. Open Settings and go to the AI Governance tab (admins only).

  2. Choose Enable AI Governance and confirm. Your people are enrolled and emailed their personal access URL automatically, with setup instructions for common tools. They can start using it straight away.

Disabling later works from the same tab; it asks you to confirm that team members have been offboarded first.

Each person's access URL

Every person gets a personal access URL, managed from the AI tab on their profile. From there you can copy the URL, resend the setup email, or rotate it. Rotating invalidates the old URL immediately, so the person needs to update their tools with the new one. The URL is what ties requests to a person, which is why it's per-person and not shared.

Connecting a tool is one configuration change: set the tool's base URL to the access URL. The setup email links each person to per-tool instructions covering Claude Code, Cursor, Windsurf, Continue, Cline, Zed, Aider, Codex CLI and the OpenAI SDKs, each with an undo step. Most tools read settings at startup, so a full restart after the change is the step people forget.

What gets metered

Each request records:

CapturedFeeds
Cost and tokens (including cache reads and writes)Spend reports and the ledger
Model and providerModel breakdown, cache and health reports
Latency and error statusAPI health reporting
Person identityAttribution: spend resolves to a person

Requests group into sessions (a stretch of AI work by one person) and sessions are what attach to tickets.

How sessions attach to tickets

Attribution uses the strongest signal available and abstains rather than guesses:

  1. Explicit ticket references in the session: authoritative when present.
  2. Nearby code activity: a commit, merged pull request or branch name by the same person within a few hours of the session that links to exactly one ticket. If the signal is ambiguous, no link is made. Unattributed is preferred over wrong.
  3. Worklog overlap: as a last resort, logged time spanning the session.

Attribution leans on the same foundation as everything else in EngLedger: people and identities. Spend from an access URL always resolves to its person; ticket references in commits and branches are what lift spend from "whose" to "what for". See AI spend attribution for how attributed and unattributed spend are reported.

What appears once data flows

Usage typically appears within minutes of a person's first request; syncs then run on a short cycle, so reports stay at most a few minutes behind. The AI Governance report suite lights up as data arrives: period spend and forecast, the AI Tax, per-ticket spend lookup, model breakdown, cache ROI, adoption, and an API Health report showing per-model error rates and latency. That's the place to check that metering is flowing.

Next steps