callhome what local AI tools send out by default, and the switch for each

the phone-home register

vLLM

What vLLM sends out of your computer by default, when, to whom, and the switch that stops it, version by version, with how we know.

The answer

By its source at v0.30.0 and its usage-stats page, not run: vLLM sends a report to stats.vllm.ai once at start (a random per-process ID; the cloud provider if detected, processor, memory, operating system, kernel, graphics cards and CUDA version; the vLLM version and the model's architecture as a class name; twenty-six run settings and five vLLM environment variables), then every 10 minutes the same ID and the time; VLLM_NO_USAGE_STATS=1, set before it starts (or DO_NOT_TRACK=1, or a do_not_track file in .config/vllm under your home folder), stops it, and we have not published a test of it.

on the research hub: A short history of vLLM · Ollama or vLLM: which to run, and how

Builds and versions

not installed: source read at tag v0.30.0On by default?yes · read in code and in its own documentsMeasured here?no · read in codeDate checked (UTC)2026-10-10
Version
0.30.0 · the source at tag v0.30.0
Where the evidence came from
not run: the source applies to Linux, the platform vLLM supports
On by default?
yes · read in code and in its own documents
Measured here?
no · read in code
Date checked (UTC)
2026-10-10
Why we say so
vllm/usage/usage_lib.py, vllm/envs.py and docs/usage/usage_stats.md read at tag v0.30.0. The source at the tag; we did not check that the released wheel matches it.

Each call out

usage report · 0.30.0

Version · build
0.30.0 · the source at tag v0.30.0 · not run: the source applies to Linux, the platform vLLM supports · not installed: source read at tag v0.30.0
Sends by default
yes · read in code and in its own documents
Kind
reports
Trigger
on its own
When
The code at v0.30.0 sends the report once at start, then every 10 minutes the same ID and the time; we did not run this version.
Where
vendor telemetry collector · stats.vllm.ai
What it sends
a random ID, new for each process (identifier); the cloud provider if detected, processor, memory, operating system and kernel, graphics cards, CUDA version (hardware and system); the vLLM version and the model's architecture as a class name (not the model's name) (software); twenty-six settings of the run (the data type, cache and memory settings, quantization, feature flags such as LoRA and prefix caching, the parallel sizes, the batching limits, the attention and compile choices, speculative decoding and the expert counts) and five vLLM environment variables (configuration)
The vendor's stated purpose
“vLLM collects anonymous usage data by default to help the engineering team better understand which hardware and model configurations are widely used.” — docs/usage/usage_stats.md at v0.30.0, read 2026-10-10
Also know
Started from a Hugging Face model ID, vLLM also reaches Hugging Face for the weights; that download is a separate call you ask for. — research.strata2signal.com · a-short-history-of-vllm, read 2026-10-02
Grade and method
read · vllm/usage/usage_lib.py, vllm/envs.py and docs/usage/usage_stats.md read at tag v0.30.0. The source at the tag; we did not check that the released wheel matches it. · read 2026-10-10
Receipts
the default server address · the switches and the do_not_track file · the 10-minute heartbeat · each report also written to usage_stats.json · its usage-stats page · a random ID, new for each process · the five environment variables it collects · the twenty-six settings passed with the report · vLLM history, 'Who can talk to it' a house page; the receipts are above

The switch

VLLM_NO_USAGE_STATS=1

Accepted values
1
What it is
environment variable of the server process
Where it must be set
in the environment of the program, before it starts (see Turn them off) · turn them off
Other spellings
DO_NOT_TRACK=1, VLLM_DO_NOT_TRACK=1, an empty file named do_not_track in .config/vllm under the home folder
What turning it off costs
none known: its documents name no other effect
Proof
found in code and documented; we have not published a test of it
For the call
usage report · 0.30.0

Listens beyond the machine, and other cloud features

Listens beyond the machine by default

  • every interface · it listens on every IPv4 address by default (cli_args.py at v0.30.0: host unset); --host localhost keeps it on the computer; our page 'Ollama or vLLM: which to run, and how' (research hub, 2026-10-08), 'Who can reach it' (0.30.0 the source at tag v0.30.0)

Other cloud features

None found.

What it leaves on disk

  • .config/vllm/usage_stats.json · a copy of what it sent while the report was on

Has it already run on your machine?

Brings in, and installed for you by

Brings in

Credit

  • The report uses a random ID that is new for each process, and the page that documents it names three ways to turn it off. — github.com · usage_stats.md, read 2026-10-10

Vendor documents it? · public discussion

Vendor documents it?

  • usage report · 0.30.0: yes · Its usage-stats page says the report is on by default, links the code that builds it, and names the switches. · github.com · usage_stats.md · read 2026-10-10

Public discussion

History

No dated changes yet.

Not yet checked

Nothing named.