OpenRouter
Usage and charges by day, model and provider. The last 30 days. AI calls, input, output, reasoning and cached tokens, the model version, workspace and the charge. Never a prompt or a reply.
Access
- Kind
- A read-only key you paste
- What to give
- Management key
- Why that one
- An OpenRouter management key. Valistry calls one read, the activity report, and nothing else. Tested first, then stored encrypted; shown only as its last four characters.
What it attributes on its own
- From the source
- day; model; the model version served; upstream provider and endpoint; workspace; requests; prompt, completion, reasoning and cached tokens; what the gateway charged, and what your own provider keys were charged
- Needs a rule or a trace
- team, workflow and agent — a gateway key carries no caller identity unless you route one key per service; outcome
The addresses it calls
- https://openrouter.ai/api/v1/activity
- https://openrouter.ai/api/v1/key
Valistry calls only these addresses. It blocks any other call and logs every call.
Connecting it
- Create a management key. Pick your organization, if any (top left) → Management Keys → New Key Open ↗
A gateway key carries no caller identity. Use one key per service, named after it, so each cost has an owner. Or add a rule in Attribution.
Settings worth turning on in OpenRouter
- One API key per app. Name each key for its app. Activity then splits by key.
- Set a limit per key. A key stops at its limit, reset daily, weekly or monthly.
- Set a guardrail budget. Calls past a member or workspace budget are rejected.
Questions about OpenRouter
What access does Valistry need?A management key. It can also manage keys; Valistry calls one read only.
How far back does it read?The last 30 full days. Older days show as missing, not zero.
How is spend split?Each day's spend, per model and provider.