Loading…
Dashboard
Live server metrics.
Active inferences ?
0
0 emb.
0 comp.
Active requests
0
Active jobs
0
OCR processes
0
Queue depth
0
Loaded models
0
Total requests ?
0
60s0/s
5m0/s
1h0/s
System resources
CPU (host) ?
0%
100%
0%
60s
now
LM-Kit server memory ?
-
-
0 B
60s
now
System memory
-
-
0 B
60s
now
GPU activity
Downloads in progress
Loaded models ?
File storage
Uploaded files
0
Total size
-
Disk free ?
-
Disk total ?
-
-
TLS status
Checking TLS certificate…
Server controls
Telemetry
Instruments--
Active-updated last 5s
Tag combinations-across all instruments
Measurements / s-rolling 5s
esc
Select an instrument
Pick one from the list to inspect its live trend, statistical distribution, and per-tag breakdown.
Alerts
Threshold breaches recorded by the server (CPU, queue depth, VRAM).
No alerts.
Scheduler
Recurring maintenance tasks. Cadence is set in appsettings.json and applied at startup; click Run now to trigger an immediate pass out-of-band.
No jobs registered.
Requests
Audit trail of recent API calls. Every row carries a request id (echoed back to the caller via X-Request-Id) so a row can be correlated with the logs and the caller's trace.
…
Audit log
… Page of …
No saved views yet.
Loading requests…
Filtering…
Logs
Server log output.
Show
Capture
?
Loading…
Loading…
Filtering…
Vector Stores
Manage vector store clusters and their collections.
Loading…
Full-Text Stores
Manage full-text store clusters and their collections.
Loading…
Search (Managed)
Manage multi-tenant search clusters. The database is created automatically if it does not exist, and the schema is initialized on create.
Search telemetry
Collecting live metrics...
Search latency p95
--
Searches / sec
--
Vector latency p95
--
DB cache hit
--
DB disk reads / sec
--
Transactions / sec
--
Longest query
--
Active DB conns
--
Conn gate queued
--
Embedding lane
--
Vector write commit
--
Backlog documents
--
Slow queries ?
Recent slow queries (live trace)
Nothing captured: no query has been observed running 3 seconds or longer.
Top queries by total database time
Loading...
Capture failed inputs
Off by default. When on, an indexing input that fails to process is saved under a per-error-code subdirectory, for offline diagnostics. Captured files may contain sensitive content.
Capture folder:
Reindex parallelism
How many documents the reindex worker processes in parallel during a rebuild, for both re-embedding (vector) and re-tokenizing (full-text). Full-text rebuild is CPU- and database-bound, so it scales with this; embedding also stays capped by the server's max concurrent inferences. Changing it takes effect immediately.
Semantic quality gate ?
The server-wide default every tenant inherits; a tenant can override it in its Tenant settings. Applies to semantic (vector) indexing of newly indexed or re-embedded documents only. Changing it takes effect immediately.
Ingestion page-size limits
Bounds how much content one indexed document can contribute, so an oversized input cannot exhaust embedding or storage. Paginated documents (e.g. PDFs) trim each page beyond the per-page limit; non-paginated documents (e.g. text, Markdown, HTML) are split into pages of the per-page limit; every document is capped at the maximum page count. Changes apply to the next indexed document.
Loading…