Integrations

Gemini CLI spend per developer

In short: turn on telemetry in ~/.gemini/settings.json with FiscaLLM as the OTLP endpoint. Each Gemini API response becomes a cost event for the developer, next to your other AI tools.

Beta. This source is built and tested against the format Gemini CLI publishes, but has not yet been run against a live account. If something looks off, open an issue.

Set it up

Create an app in FiscaLLM (free) and copy its ingest key. The app's Sources page shows these settings with your key filled in.

In ~/.gemini/settings.json:

{
  "telemetry": {
    "enabled": true,
    "target": "local",
    "otlpProtocol": "http",
    "otlpEndpoint": "https://fiscallm.com/api/v1/otel",
    "otlpHeaders": { "Authorization": "Bearer fsk_your_key" },
    "logPrompts": false
  }
}

What is read

Limits

Using more than one AI tool? Each one is a source you turn on in the same app, and the overview shows the cost of each tool, developer and model side by side. See all integrations.

Questions

Do I need an OpenTelemetry Collector or Google Cloud?

No. With "target": "local" and "otlpProtocol": "http", Gemini CLI posts its telemetry straight to FiscaLLM. Nothing goes to Google Cloud.

How is the cost worked out?

From the token counts of each API response: uncached input, cached content, output and thinking tokens, at Google's API list prices for the model.

Are prompts sent?

Set "logPrompts": false, as in the settings above. Even when prompts are sent, FiscaLLM reads only the token counts of the gemini_cli.api_response event.

Is anything counted twice?

No. Gemini CLI reports each call twice, once as gemini_cli.api_response and once in the GenAI conventions; only the first is read.