Installation

llmprof is a small Python package. It needs Python 3.9+ and runs entirely on your machine.

pipx (recommended)

pipx installs the llmprof CLI into its own isolated environment so it never clashes with your project’s dependencies.

pipx install llmprof
llmprof up

npx (no Python needed)

Not a Python shop? Run llmprof straight from npm - no Python, pip, or virtualenv required:

npx llmprof up

The launcher bootstraps uv (a single static binary that provisions its own Python) and runs the real llmprof package. The first run downloads uv once and caches it under ~/.cache/llmprof; every argument is forwarded, so npx llmprof traces and npx llmprof up --port 4100 work as expected.

Useful environment overrides:

• LLMPROF_SPEC - the package spec to run (default llmprof==<launcher version>); point it at a path or git+https://... URL for a dev build.
• LLMPROF_CACHE_DIR - where the uv binary is cached.
• LLMPROF_UV - path to an existing uv binary to use directly.

pip

If you would rather install into the current environment (for example to use the Python SDK alongside your app):

pip install llmprof

From source

While the project is pre-launch, or if you want to hack on it:

git clone https://github.com/luthraG/llmprof
cd llmprof
pip install -e .
llmprof up

Docker

docker compose up

This serves the proxy and dashboard on port 4000 and keeps the trace database in a mounted volume so your history survives restarts.

Verify

With the proxy running, check it is healthy and open the dashboard:

curl http://localhost:4000/llmprof/health
# {"ok": true, "upstream": "https://api.openai.com"}

Open http://localhost:4000 in a browser. It will be empty until you send a call - head to the Quickstart.

What gets stored, and where

llmprof writes captured traces to a single local SQLite file at ~/.llmprof/llmprof.db (override the directory with LLMPROF_HOME). Nothing is sent anywhere except the upstream provider you are already calling. See Storage backends to point it at a shared database instead.