pAIring stats and monitoring
Admins open Stats from the sidebar. It has five views.
| View | What it shows |
|---|---|
| Overview | Who uses pAIring and how much |
| Visual chat workflow | How a message travels through the system |
| Telemetry | Every call pAIring makes to the engine |
| System | CPU, memory, disk and GPU |
| Supported architectures | What Matricxon can load (Matricxon only) |
Overview
Section titled “Overview”A usage summary:
- Conversations: total, private chats and channels, and the count per model.
- Messages: total, by role (user or assistant) and by status, such as failed or stopped.
- Users: total, active and disabled, and the count per role.
- Knowledge base: documents, chunks, and shared versus private documents.
- Live conversations: how many are streaming a reply right now, and how many were active recently.
Visual chat workflow
Section titled “Visual chat workflow”An interactive diagram of the path a message takes, from the browser through pAIring to the engine and back. Each box is colored by layer, as the legend in the corner shows. Click a box to read what that step does. The diagram scrolls in both directions.
Telemetry
Section titled “Telemetry”Every chat and embedding call pAIring sends to the engine is recorded with OpenTelemetry. It is call-level data, not a list of who is online.
- Total requests and errors.
- Requests per model and per host.
- Average latency per model, and latency over time.
- Currently loaded models, with their size in RAM and VRAM and when they unload. This is checked every 30 seconds.
A selector picks which engine’s history to show. It starts on the active engine. Switching it does not change which engine answers chats.
System
Section titled “System”Live readings of the pAIring server itself, taken every 15 seconds:
- CPU and memory use, with a chart over time.
- Disk use.
- GPU: utilization, memory and temperature per NVIDIA card, read with
nvidia-smi. On a Mac, overall memory is shown for the built-in GPU.
If no GPU is found, the GPU section says so.
Supported architectures
Section titled “Supported architectures”Shown only when Matricxon is the active engine. It lists the model architectures and quantization types the running Matricxon can load. See Supported models.
How long data is kept
Section titled “How long data is kept”Telemetry and System history are cleaned every 6 hours. By default, Telemetry keeps 30 days and System keeps 3 days. Change both under Admin settings → System → Data retention.