Usage statistics

SnowLLM reports usage statistics by default. Reports only include a one-time session ID, software version, OS, GPU architectures, model class, and usage counters; they exclude model IDs, Recipe IDs, and conversation content.

View usage statistics →

Collected fields

Field	Meaning
session_uuid	In-memory UUIDv4; regenerated on restart.
snowllm_version	SnowLLM version.
os	linux, wsl, windows, or macos.
gpu_arch	AMD GPU architectures separated by ;, preserving duplicates; null if unrecognized.
uptime_s	Seconds since service startup.
load_count	Successful model loads this session.
model_class	Current model architecture class; null if unloaded.
prefill_tokens	Cumulative prefill tokens since loading, excluding prefix-cache hits.
decode_tokens	Cumulative accepted decode tokens since loading, excluding discarded speculative tokens.

Reporting

Reports are sent via HTTPS JSON POST to https://status.snowllm.dev/ at startup, every 10 minutes, after each successful model load, and before unloading.

Telemetry source code

Disable telemetry

Environment variable

Set before starting the service:

Linux / WSL / macOS

export SNOWLLM_NO_USAGE_STATS=1

Windows

$env:SNOWLLM_NO_USAGE_STATS = "1"

DO_NOT_TRACK=1 also works.

Marker file

Alternatively, create an empty do_not_track file before starting the service:

Configuration	File location
Linux / WSL / macOS	${XDG_CONFIG_HOME:-$HOME/.config}/snowllm/do_not_track
Windows	%APPDATA%/SnowLLM/do_not_track
Custom configuration	Beside the config file selected by SNOWLLM_CONFIG or the daemon’s --config

All systems also check ~/.config/snowllm/do_not_track. Default locations remain effective with custom configurations.