Keiko 0.3.15 — activity log

New

An activity log you can read.

Keiko now writes one structured line per operation to logs/server.log inside its state directory
— active from the moment it starts, no setting to switch on.

{"ts":"…","category":"indexing","op":"indexing.job.started","correlationId":"9c111343-…","extra":{"batchSize":64,"concurrency":4}}
{"ts":"…","category":"embedding","op":"embedding.batch.dispatch","correlationId":"9c111343-…","extra":{"inputCount":36,"bodyBytes":48211,"timeoutMs":30000}}
{"ts":"…","category":"http","op":"http.gateway.fetch.completed","correlationId":"9c111343-…","status":500,"durationMs":19003}
{"ts":"…","category":"embedding","op":"embedding.batch.degrading-to-scalar","correlationId":"9c111343-…","errorKind":"http-error","status":500}
{"ts":"…","category":"indexing","op":"indexing.job.finished","correlationId":"9c111343-…","extra":{"vectorsPersisted":36,"status":"succeeded"}}

• One run, one identifier. Every line of an indexing run carries the same job id, so
  acceptance, document discovery, extraction, chunking, embedding requests, retries, fallbacks and
  the final outcome are followed with a single search of one file.
• In-flight calls are visible. Outbound gateway calls are recorded before they are sent and
  again when they answer, with the item count, byte size, configured timeout, elapsed time and HTTP
  status. A request that has not returned shows as such, with how long it has been running.
• Everything that can fail says so. Retries and their backoff, every rung of a compatibility
  fallback, circuit-breaker transitions, egress and TLS failures, store quarantine, and each of the
  refusal paths a request can take.

Privacy

The log is content-free by construction, not by convention:

• Counts, sizes, durations, HTTP statuses, error classes, and identifiers reduced to digests.
• Document text, model output, credentials, request bodies, absolute file paths and names you chose
  yourself cannot reach it — request paths are recorded as route templates, so
  /api/local-knowledge/capsules/{id}/index rather than the pod name you typed.

The file can be handed to support exactly as it is.

Reading it

grep '<job-id>' <keiko-dir>/.keiko/logs/server.log

The job id is shown in the pod view. Rotation is daily, seven days are kept.

Verbosity is adjustable through KEIKO_LOG_LEVEL (debug, info, warn, error). The default,
info, records the course of every operation without per-item detail.

Operator action

Restart the Keiko process after updating — the running process keeps serving the old code even once
the footer reports 0.3.15. No gateway re-setup and no configuration change are required.

npm install -g @oscharko-dev/keiko@0.3.15