Telemetry configuration

Browser bindings and the native CLI emit a terminal record when a high-level inference call returns. Records describe the model, runtime, execution, and coarse platform identity.

Configure the browser#

The default receiver is https://telemetry.minifieldlabs.com/insert/jsonline. Configure the WASM module in the Worker that loads and runs it.

import init, {
  configure_telemetry,
  flush_telemetry,
} from './runtime/minifield_web_demo.js';

await init();
configure_telemetry({
  integrationId: 'contact-extraction',
  productId: 'contact-editor',
  applicationId: 'web-app',
  applicationVersion: '1.0.0',
  environment: 'production',
});

// Flush queued records before terminating this Worker.
await flush_telemetry();

Use preview, development, or test for those environments. Labels accept up to 128 ASCII letters, digits, or ._:/@+-.

Disable delivery#

configure_telemetry({ enabled: false });

For the native CLI:

MINIFIELD_TELEMETRY=0 minifield-infer \
  --model-dir /absolute/model \
  --bos true \
  --max-output-tokens 12 \
  --max-context-tokens 64

Native library hosts choose reporting through run_with_reporter. run_with_io performs no network I/O.

Use your receiver#

configure_telemetry({
  enabled: true,
  endpoint: 'https://metrics.example.com/insert/jsonline',
});

Use MINIFIELD_TELEMETRY_ENDPOINT for the native CLI. Receivers require HTTPS, with loopback HTTP accepted for local tests.

Record contents#

Schema minifield.runtime-inference/1 records elapsed duration, token counts, estimated arithmetic work, backend, weight formats, model fingerprint, and deployment labels.

Prompt text, generated text, tool arguments, and error messages stay outside the record. Browser origin contains the scheme, host, and port. Delivery omits credentials and referrer.

See the complete runtime telemetry schema for field definitions and batching behavior.