CanalAPI
User GuideDeveloper GuideExamples

Usage Logs

Inspect per-request token usage, cost, and errors in the console; export CSV for offline analysis.

The Usage Logs panel records every request that hits CanalAPI. It is the primary tool for monitoring activity and debugging unexpected charges or failures. Logs are retained for 30 days by default.

Where to find it

Open Console → Usage Logs. By default the most recent requests are shown first.

Available fields

Each log entry typically includes:

FieldNotes
TimestampWhen the request was processed
API Key nameThe key used for the request (the secret is never shown)
ModelThe model the request was routed to
Input tokensToken count of the prompt portion
Output tokensToken count of the completion portion
CostAmount deducted, in the account's display currency
LatencyServer-side processing latency in milliseconds
StatusSuccess / failure; failed rows include error code and a short message
Request IDIdentifier used in support tickets and programmatic reconciliation

Click any row to open the detail drawer with:

  • Full request ID (one-click copy)
  • Endpoint path and model
  • Input / output tokens, cost, latency
  • Error code and short summary on failure
  • Whether streaming was used

Prompt / completion content is never shown — CanalAPI does not retain request / response bodies. See Privacy below.

Filtering

The toolbar provides these filter dimensions:

  • Time range — Today / Yesterday / Last 7 / 30 / 90 days / This month / Last month / This quarter / This year / Last 2 years / All, or a custom range
  • API key — single or all
  • Model — single or all, with search
  • Status — All / Success / Failed / Errors
  • Keyword — exact match on request ID and similar fields

Filter state is persisted in the URL for easy sharing.

Export

Click Export CSV to stream out the rows matching your current filters.

  • Large result sets stream in chunks; no need to wait for a fully-built file.
  • Exports are rate-limited per account (5 per hour by default); excess requests return 429.
  • Column order matches the list view, so the file imports cleanly into Excel / BigQuery / Snowflake.

For deeper reconciliation, combine with the OpenAI-compatible GET /v1/dashboard/billing/credit_grants balance endpoint.

Dashboard analytics

Console → Dashboard aggregates the same data into:

  • Today / this-month spend with previous-period comparison
  • Cost breakdown by model / key
  • Usage trend (per hour / day)
  • Per-key usage trend comparison

Use it to forecast spend, spot anomalies, and attribute cost.

Troubleshooting

Failed requests appear in usage logs as Status = Failed. Open the detail drawer to see:

  • HTTP status code and platform error code
  • A short upstream message (sanitized) where applicable
  • Whether the request was billed (most failures are not; some upstream failures after generation begins may still be)

Full error catalog and handling guidance: Errors.

Privacy

User-facing usage logs show metadata only (tokens, model, timestamp, status, error code):

  • No full prompt / completion content
  • No tool-call arguments or results
  • No upstream API keys

For the authoritative description of what is collected and retained, see the Privacy Policy in the console.

On this page