Log

What it is

Log records every call to /v1/*: caller, model, token consumption, channel hit, status code, error message. All traces are persisted for troubleshooting and auditing.

Where to find it

• Admin → Logs: filter by time / user / model / status; paged
• Personal Center → My Logs: the current user's own logs (compact view)
• Admin → Dashboard: today's / this month's stats

Operator-relevant fields

Field	Meaning	Effect
Caller	User	"Who is calling".
Token	Token FK	"Which key is calling".
Channel	Channel FK	"Which upstream path was hit".
Model	e.g. gpt-4o	"Which model was used".
Input / output tokens	Number	Compute cost, locate overruns.
Status code	200 / 401 / 429 / 5xx	First place to look on failure.
Error message	Text	Raw error from the upstream provider.
Latency	ms	Locate slow requests.
Quota charged	Number	Final deduction for this call.

Common troubleshooting paths

• User reports "charged too much": check input/output tokens vs the upstream provider's own billing; ensure token counting matches.
• User reports "always 429": look for "quota insufficient" in the error message; check User.quota.
• User reports "model not working": see if channel is empty; check whether the model is selected in the channel.
• Locating slow requests: sort by latency descending; the slowest channel is the suspect.

Cleanup / archival

Logs grow unbounded. Configure a retention window (90 days is typical in production). Admin → Logs → "Clear history" deletes logs older than the threshold.

Related pages

• Logs (admin)

Related API

• GET /api/log/ — list (admin)
• GET /api/log/self — current user's logs
• GET /api/log/stat — stats
• DELETE /api/log/ — clear history