Getting Started

Install, connect data, connect AI, and ask the first question.

This guide walks through the shortest self-service path for Vnuli 0.1: install the package, initialize a project, register supported engineering data, connect a qualified AI client, and ask for deterministic evidence.

1. Install Vnuli

Install Vnuli from PyPI in the Python environment you want to use.

pip install vnuli

Verify that the command is available.

vnuli --version
vnuli --help

2. Initialize a project

Run setup from the folder where you want Vnuli project state to live. Vnuli creates its own local project state; you should not edit it by hand.

vnuli setup

3. Connect a simple CSV dataset

The easiest first dataset is the supported CSV layout: a UTF-8 file with a header row, a first column named exactly time, and numeric channel columns. Create a file named first-test.csv in your project folder:

time,pressure,temperature
0.0,1.0,70.0
0.1,2.0,70.5
0.2,3.0,71.0
0.3,4.0,71.5

Register and inspect it.

vnuli data add first-test.csv
vnuli data list
vnuli data show csv_first_test

For this file, Vnuli registers dataset csv_first_test, run run_001, one explicit timestamp timebase, and channels pressure and temperature. CSV units are reported as unknown; Vnuli does not infer units from column names.

For HDF5, MATLAB MAT v5, TDMS, and NASA Milling reference data, see Supported Data.

4. Connect a qualified AI client

Pick a qualified client you use. This example uses Codex; GitHub Copilot and Gemini CLI are also qualified in tested environments.

vnuli mcp install codex
vnuli mcp list
vnuli mcp verify codex

These commands configure the client to launch the same local Vnuli MCP server, vnuli serve. Configuration verification confirms the local setup; the real product check is that your AI client can discover Vnuli tools and use them on the registered dataset.

Use Supported AI Clients for exact client status, tested environments, and limitations. Claude Code is currently deferred and should not be treated as qualified.

5. Ask the first engineering question

Open the qualified AI client from the project folder and ask a natural question like this:

Use Vnuli tools to inspect the registered dataset. What runs and channels are available, and what are the basic statistics for the pressure channel?

The client should use Vnuli tools such as run/channel discovery, metadata retrieval, and statistics. For the CSV above, the pressure values are 1.0, 2.0, 3.0, 4.0, so the deterministic mean is 2.5, minimum is 1.0, maximum is 4.0, and range is 3.0.

6. Inspect evidence and provenance

In the result, look for the source dataset, run, channel, method, analysis window, warnings, and provenance. The AI is not the numerical authority. Vnuli performs the deterministic calculation and returns bounded evidence the AI can explain.

7. Ask a follow-up

Continue the investigation with a second question:

Now use Vnuli to show the pressure time history over the full recorded window and summarize how it changes over time.

This keeps the investigation grounded in Vnuli tool results while the AI helps navigate and explain the evidence.

Where to go next