Observability & usage

Every request you make is recorded with its model, protocol, token counts, latency, error (if any), and the exact credit charged. The console gives you both per-request logs and aggregated usage views.

Logs vs. usage

Logs

Per-request detail: the model, protocol, token usage, latency, status, and the credit deducted. Open a record to inspect a single request.

Usage

Aggregated views: totals over time, top models, top API keys, and stacked breakdowns by model, key, or token type.

Use Logs to debug a specific request and Usage to understand spend and volume over a date range.

What's recorded

For each request you can see:

The model slug you requested and the protocol path it took.
Input, output, and cache token counts.
Latency — both time to first token (for streams) and total duration.
The credit charged, computed from OneHop's pricing engine.
The error code, if the request failed.

Latency is observation-only

Timing fields like first-token and total duration are measured from the moment OneHop forwards upstream. They are for monitoring and troubleshooting only and never affect what you're billed.

Failed requests

Failed requests are recorded with a specific error code so you can tell what went wrong:

upstream_timeout — the upstream or stream timed out.
client_disconnected — the client disconnected before the upstream responded.
upstream_unreachable — a network or DNS-level failure.
usage_unsettled — no settleable usage after the reconciliation window.

Low-balance alerts

Set a low-balance threshold in the console to get notified before your wallet runs dry. The default threshold is conservative, so even accounts living on bonus credit get a heads-up before requests start failing. See Billing & wallet for how charges draw down the wallet.

Agent-friendly docs

Building an agent on top of OneHop? Every page in these docs is also available as raw Markdown — append .md to any doc URL to get a clean plaintext view your model can read directly.