Metadata-Version: 2.4
Name: vnuli
Version: 0.1.5
Summary: Vnuli tools for AI-assisted physical engineering investigation.
Requires-Python: >=3.12
Description-Content-Type: text/markdown
Requires-Dist: h5py>=3.11
Requires-Dist: matplotlib>=3.8
Requires-Dist: mcp[cli]>=2
Requires-Dist: nptdms>=1.10
Requires-Dist: numpy>=1.26
Requires-Dist: scipy>=1.12
Provides-Extra: test
Requires-Dist: pytest>=8; extra == "test"

# Vnuli

**AI tools for engineers of the physical world.**

Vnuli lets engineers use the AI client they already work with to investigate
engineering data through deterministic tools. The AI can ask Vnuli to inspect
registered data, run calculations, and return bounded evidence with provenance.

```text
Connect engineering data
-> use a qualified AI client
-> Vnuli performs deterministic engineering analysis
-> receive bounded, traceable evidence and provenance
-> continue investigating through follow-up questions
```

The AI is not the numerical authority. Vnuli performs the engineering
computation and returns evidence the AI can use to explain results and guide
the next question.

## Install

Vnuli 0.1 is qualified on Windows 11 x86-64 with Python 3.12.

```powershell
pip install vnuli
```

Verify the install:

```powershell
vnuli --version
vnuli --help
```

## First Use

From the project folder where you want Vnuli state to live:

```powershell
vnuli setup
vnuli data add <source>
vnuli data list
vnuli data show <dataset>
vnuli mcp install <client>
vnuli mcp verify <client>
```

What those steps do:

- `vnuli setup` initializes Vnuli project state.
- `vnuli data add <source>` registers a supported engineering data source.
- `vnuli data list` shows registered datasets.
- `vnuli data show <dataset>` shows Vnuli's interpretation of a dataset.
- `vnuli mcp install <client>` configures a qualified AI client to launch
  Vnuli's local MCP server.
- `vnuli mcp verify <client>` verifies the local client configuration.

All qualified AI clients use the same local Vnuli server:

```powershell
vnuli serve
```

Normal setup does not require hand-authored MCP JSON/TOML.

## Supported Data

Vnuli supports documented engineering time-series layouts, not arbitrary files
with matching extensions.

Current qualified data categories:

- CSV
- HDF5
- MATLAB MAT v5
- TDMS
- NASA Milling `mill.mat` reference data

See [Vnuli Supported Data Layouts](docs/vnuli-supported-data-layouts.md) for
the exact layouts, timebase behavior, metadata behavior, bounded-access notes,
and unsupported-layout rules.

## Supported AI Clients

Current AI-client compatibility status:

| Client | Status |
|---|---|
| Codex | Engineering workflow qualified / PASS |
| GitHub Copilot | Engineering workflow qualified / PASS |
| Gemini CLI | Engineering workflow qualified / PASS |
| Claude Code | Engineering workflow qualified / PASS |

See [Vnuli Supported AI Clients](docs/vnuli-supported-ai-clients.md) for tested
versions/environments, configuration locations, limitations, and qualification
details. Qualification of one tested version/environment does not certify every
future client version, account type, operating system, or extension build.

## Trust Model

Vnuli keeps engineering computation in deterministic tools:

- source data is registered and interpreted by Vnuli;
- engineering calculations are performed by Vnuli, not guessed by the AI;
- results include bounded evidence and provenance;
- the AI receives Vnuli tool results and explains or follows up from that
  evidence.

This separation is the core product boundary:

```text
AI client
-> Vnuli engineering tools
-> deterministic computation
-> bounded evidence + provenance
-> AI explanation and follow-up
```

## Learn More

- [Supported Data Layouts](docs/vnuli-supported-data-layouts.md)
- [Supported AI Clients](docs/vnuli-supported-ai-clients.md)
