See what a session cost and what a run did

On this page

CLIO Coder records what each process, session, and worker run did in local journals, and every usage figure on every surface comes from those journals. This guide shows how to read a session's cost, find out why it stopped, check a run's record, and prepare an archive for a maintainer.

Before you start

CLIO Coder installed, and at least one finished session in a repository.
The session id, which the exit summary prints. /resume also lists them.

01
Read it in the session
```
/usage
/context
/usage health
```
/usage covers session usage and supported subscription quotas. /context shows context use and the newest call's cache evidence. /usage health shows process health and route state.
02
Read it from the shell
Shell
```
clio-coder telemetry inspect <session>
clio-coder telemetry cost <session> --by model
clio-coder telemetry calls <session> --limit 20
clio-coder telemetry cache <session>
```
Each takes --json and returns the same data the terminal, editors, and desktop app read. For a repository over time, use clio-coder telemetry cost --repo . --days 30 --by model.
03
Ask why it stopped
Shell
```
clio-coder telemetry why-stopped <session>
clio-coder telemetry timeline <session>
clio-coder telemetry explain <ref>
```
explain shows how a figure was derived. --run <run-id> points inspect and cost at one worker run instead of a session.
04
Check a record
Shell
```
clio-coder telemetry verify <session>
clio-coder telemetry verify --run <run-id>
```
A sealed scope directory is self-contained, so you can copy it and verify it on another machine. verify checks journal segments, objects, and seals.
05
Share it deliberately
Shell
```
clio-coder telemetry share --session <id> --profile content-free --dry-run
clio-coder telemetry share --session <id> --profile content-free --out report.tar.gz
```
Without --send, share writes an archive and uploads nothing. content-free is the default profile, with structural records and no objects or content. structural adds objects that hold structure, such as prompt manifests. Read the dry run before you send anything.

Terminal output: telemetry why-stopped shows a closed session with spend not measured, telemetry verify reports the session sealed with 66 lines in one segment, and telemetry cost by turn shows tokens as lower bounds and cost as not measured. — A gateway that reports no usage: token figures are lower bounds, cost reads not measured, and the record still verifies. Terminal · v0.6.8

What you should see

Figures labelled by provenance, and a verify result that names any gap. A clean verify shows the journals are internally consistent. The chain is unsigned, so it does not show who wrote them, and it says nothing about whether the work was correct.

Control what is kept

telemetry.mode defaults to standard, which keeps accountability and operational records without prompt content. research mode, telemetry.research.content: true, and clio-coder telemetry consent for the workspace are all required before prompts and request bodies are kept. Consent is stored in your own state directory, keyed by the workspace's real path, so a file committed to a repository cannot grant it.

clio-coder telemetry prune is a dry run unless you pass --apply.

Boundaries

Figures depend on what the provider reports. Calls with unknown usage are counted separately.
Quota rows are what a provider account reports, shared across sessions, and are separate from session spend.
clio-coder usage report is experimental.

Before you start

Read it in the session

Read it from the shell

Ask why it stopped

Check a record

Share it deliberately

Control what is kept