io.github.nveil-ai/nveil

NVEIL is an AI-powered data visualization toolkit. Write one line of natural language, and NVEIL processes your data and generates publication-ready visualizations — no chart code, no hallucinations, no data leaving your machine.

import nveil

nveil.configure(api_key="nveil_...")

# Pass a file path directly — no DataFrame loading required.
spec = nveil.generate_spec("Revenue by region, colored by quarter", "sales.csv")

fig = spec.render("sales.csv")   # 100% local — no API call
nveil.show(fig)                   # opens in browser

From your shell

After pip install nveil the nveil command is on your $PATH:

export NVEIL_API_KEY=nveil_...

# Ground yourself on the dataset (shape / dtypes / head preview)
nveil describe sales.csv

# Generate HTML + PNG + a reusable .nveil spec, print the explanation
nveil generate "Revenue by region, colored by quarter" \
  --data sales.csv --format all --explain

# Re-render an existing spec on fresh data — no API call
nveil render chart.nveil --data new_sales.csv

For AI agents (Claude Code / Claude Desktop / Cursor / Codex / …)

NVEIL ships first-class integrations:

# Claude Code / Claude Desktop — install the bundled skill
nveil install-skill

# Claude Desktop, Cursor, any MCP client — add an MCP server:
# {"mcpServers": {"nveil": {"command": "nveil", "args": ["mcp"]}}}
nveil mcp                    # stdio server; launched by the MCP client

Why NVEIL?

_{¹ ChatGPT Advanced Data Analysis, Claude Analysis tool, Gemini Data Agent  ·  ² PandasAI, LIDA, Julius, Vanna  ·  ³ Plotly, Matplotlib, Seaborn}

How It Works

Your Data ──> Toolkit ──metadata only──> NVEIL AI ──> Processing Plan ──> Local Execution ──> Result
               ^                                                           ^
          raw data stays here                                     raw data stays here

1. You describe what you want in plain language
2. NVEIL AI plans the data processing and visualization (only metadata is sent — column names, types, statistics)
3. The Toolkit executes locally — joins, aggregations, pivots, rendering — all on your machine
4. You get a figure — Plotly, VTK, or DeckGL, auto-selected for your data

Key Features

🧠 Two Engines in One

Data processing (joins, pivots, aggregations, geocoding, time series) AND visualization generation from a single prompt.

🔒 Data Privacy by Design

Raw data never leaves your machine. Only column names, types, and aggregate statistics are sent.

📈 Multi-Backend Rendering

Auto-detects the best engine: Plotly (2D charts), VTK (3D/medical), DeckGL (geospatial).

🧪 Auditable Results

Powered by constraint solving, not random generation. Same input = same output, every time.

⚡ Offline Rendering

spec.render() runs 100% locally with zero API calls.

💾 Reusable Specs

Save to .nveil files, reload later, render on new data — no server needed.

Beyond Simple Charts

NVEIL handles geospatial heatmaps, 3D volumes, scientific visualizations, medical imaging (DICOM), biosignal data (EDF/EDF+), network graphs, and 50+ other visualization types — all from natural language.

Save Once, Render Forever

# Generate once (API call)
spec = nveil.generate_spec("Monthly trend by category", df)
spec.save("trend.nveil")

# Reload anywhere — no API call, no server, no cost
spec = nveil.load_spec("trend.nveil")
fig = spec.render(fresh_data)
nveil.save_image(fig, "report.png")

Installation

pip install nveil

Requirements: Python 3.10+

Getting Started

1. Create an account at app.nveil.com
2. Generate an API key in Settings
3. Start visualizing

import os
import nveil

nveil.configure(api_key=os.environ["NVEIL_API_KEY"])

spec = nveil.generate_spec("scatter plot of price vs area", df)
fig = spec.render(df)
nveil.show(fig)

See the examples/ directory for more usage patterns.

Documentation

Full documentation is available at docs.nveil.com:

• Quickstart Guide
• Core Concepts — sessions, specs, and the two-stage flow
• API Reference — full reference for all public functions
• Privacy Model — what data is sent, what stays local
• Examples — bar charts, multi-dataset, offline rendering

Contributing

Contributions are welcome under the project's Contributor License Agreement. Bug reports and feature requests are welcome via GitHub Issues.

License

GNU AGPL v3 or later. See LICENSE. Commercial dual-licensing is available — contact pierre.jacquet@nveil.com.