Codex Turn-Level Usage in Beacon

What changed for Codex?

Codex already belonged in the endpoint agent record. The missing piece was granular, attributable usage. Aggregate token counts can tell a team that spend happened, but they do not explain which session, model, turn, or user produced it.

Beacon CLI v1.2.5 closes that gap by configuring Codex's local trace exporter and retaining the completed turn span that carries the thread, turn, model, and token fields.

How Beacon reads Codex usage

Beacon keeps one live source for Codex usage. The managed configuration uses local traces rather than enabling the anonymous token histogram alongside them. Historical metric rows remain readable, and post-cutover overlaps prefer the attributable turn span.

The completed turn is normalized into the same canonical gen_ai.usage schema used by Claude Code. It records uncached input, cache read, cache creation, output, and reasoning tokens.

A metadata-only SessionStart hook records the local OS user for each Codex session, so the token report and dashboard can join usage back to the session and show per-user breakdowns.

What you can do now

With this update, teams can:

• Attribute Codex token usage by completed turn, session, model, repository, and local user.
• Compare Codex and Claude Code usage through the same canonical usage schema.
• Keep rollout JSONL as an explicit recovery source without continuously scanning Codex transcript files during normal endpoint execution.