What is AI usage telemetry?
AI coding tools bill per seat but get used wildly unevenly. Telemetry is how Velory finds out who actually uses them: the tools on your team's machines report their own consumption — tokens, models, cost — and Velory attributes it to the right employee.
It answers the question a seat count cannot. Forty licences tell you what you pay for; telemetry tells you that eleven of those people have not run a prompt in a month.
In this guide
Before you begin
You need Admin access. Setting telemetry up is a fleet exercise rather than something you switch on per service — see How do I set up AI usage telemetry?.
Which tools report
Three, matched to the services in your catalogue:
| Tool | Reports as |
|---|---|
| Claude Code | Claude |
| Codex | OpenAI |
| Gemini | Gemini |
The service has to exist in your software catalogue under that exact name for the usage to land against it.
Any other software will never produce telemetry. That is a limit of what the tools themselves report, not a setting you have missed.
What it collects
Per person, per day, per model:
- The employee's work email, used to attribute the usage
- Which model was called
- Input and output tokens
- Cache read and cache creation tokens
- Reasoning tokens, where the model supports extended thinking
- Cost, where the tool reports it
It does not collect prompts or responses. No message content, no code, no file names — only counts, model names and a date. If someone asks whether Velory can read what their team is writing in Claude Code, the answer is no.
Usage is recorded to the day rather than the minute, so telemetry tells you that somebody worked on a Tuesday, not when they started.
What it powers
Telemetry is not a standalone report. Four things depend on it:
| Feature | What telemetry gives it |
|---|---|
| Health scores | The 0–100 score and A–D grade on each licence — the basis for reclaiming unused seats |
| The Telemetry tab | Per-person tokens, cost and models on a service |
| Pay Per Use spend | Accrued spend on a plan billed by consumption |
| SaaS Discovery | A signal source, so an AI tool in use shows up even if nobody registered it |
If your licences all show Not scored yet, this is why: no telemetry is arriving. See How do I find and reclaim unused licences?.
How it differs from a usage report
Two separate mechanisms, easily confused:
| Telemetry | Usage report | |
|---|---|---|
| Where it comes from | The tool on each machine | The vendor's own account records |
| Set up | Once, across the fleet, via MDM | Per service |
| Covers | Claude Code, Codex, Gemini | Claude, OpenAI |
| Produces | Health scores, tokens, cost, discovery signal | The License usage sub-tab |
| Can reclaim seats | No | Yes, on the CSV path |
They are complementary rather than alternatives. Telemetry tells you how hard a person is using a tool; a usage report is the vendor's account-level view. See How do I import a usage report?.
Good to know
Does this need a Velory agent on people's machines?
No. The tools already emit their own usage in a standard format; the configuration you distribute points them at Velory instead of nowhere. There is nothing of Velory's to install.
Whose usage shows up?
Anyone whose machine got the configuration. Attribution is by work email, so somebody signed in with an address that does not match their employee record shows as Unmatched rather than being tied to a person.
What happens to someone's usage after they leave?
It moves into an anonymised Former employees bucket. The totals stay honest — the cost was real — but it is no longer attributed to a named person.
Is the cost figure what we are billed?
No. It is an approximation at API-equivalent pricing, which is exactly right for comparing people against each other and wrong for reconciling an invoice. A seat-based or monthly plan bills differently regardless of what was consumed.
Can I turn it off?
Yes, by revoking the telemetry token or removing the configuration from your fleet. Reporting stops; data already collected stays.
Was this article helpful?