Venturalítica AI Assurance SDK

AI Assurance through Compliance-as-Code.

Venturalítica transforms your Python code into Legal Evidence. It automatically maps your technical metrics, data audits, and execution logs to the EU AI Act (Articles 9-15) without leaving your local environment.

Open Source & Free. Help us build the standard for AI compliance in Europe.

⭐ Star on GitHub 📦 View on PyPI

Quickstart in 60 Seconds

pip install venturalitica

pip install "venturalitica[metrics]"

pip install "venturalitica[dashboard]"

pip install "venturalitica[agentic]"

pip install "venturalitica[full]"

Detect bias in your datasets or models with one line of code:

import venturalitica as vl

# Run a full AI Assurance audit on the built-in loan scenario
results = vl.quickstart('loan')

Then launch the AI Assurance Dashboard to explore results:

venturalitica ui

Key Features

Feature	Description
`enforce()`	Audit datasets and models against OSCAL policies with 35+ built-in metrics.
`monitor()`	Wrap training runs with automatic evidence collection (BOM, hardware, carbon).
Glass Box Dashboard	4-phase regulatory workflow: Identity, Policy, Verify, Report.
Policy as Code	Define assurance rules in OSCAL `assessment-plan` format.
Column Binding	Decouple policies from schemas via synonym-based column resolution.
Local Sovereignty	Zero-cloud dependency. All enforcement runs locally.
Annex IV	Auto-draft technical documentation from local traces.

Architecture

┌─────────────────────────────────────────────────────────────┐
│                     Your ML Pipeline                        │
│                                                             │
│   DataFrame ──► vl.enforce(policy=…) ──► ComplianceResult[] │
│                        │                                    │
│                  OSCAL Policy                               │
│               (assessment-plan)                             │
└───────────┬─────────────────────────────────┬───────────────┘
            │                                 │
     ┌──────▼──────┐                   ┌──────▼──────┐
     │  35+ Metrics │                  │  7 Probes   │
     │  ─────────── │                  │  ─────────  │
     │  Fairness    │                  │  Hardware   │
     │  Data Quality│                  │  Carbon     │
     │  Performance │                  │  BOM        │
     │  Privacy     │                  │  Artifact   │
     │  Robustness  │                  │  Trace      │
     └──────┬──────┘                   └──────┬──────┘
            │                                 │
            └───────────────┬─────────────────┘
                            │
                     ┌──────▼──────┐
                     │  Dashboard  │
                     │  ─────────  │
                     │  Identity   │
                     │  Policy     │
                     │  Verify     │
                     │  Report     │
                     └─────────────┘

Documentation

Guide	Description
Quickstart	Run a full compliance scan in 2 minutes.
API Reference	`enforce()`, `monitor()`, `wrap()`, `quickstart()`, `PolicyManager`.
Metrics Reference	All 35+ metrics across 7 categories.
Policy Authoring	Write OSCAL policies from scratch.
Dashboard Guide	The Glass Box 4-phase workflow.
Column Binding	Map abstract names to DataFrame columns.
Probes Reference	7 evidence probes for EU AI Act compliance.
Experimental Features	CLI `login`/`pull`/`push` (SaaS preview).

Academy (Step-by-Step Learning)

Level	Role	Focus
Level 1	Policy Author	Write your first OSCAL policy for the loan scenario.
Level 2	Integrator	Add `enforce()` to your ML pipeline.
Level 3	Auditor	Review evidence and interpret results.
Level 4	Compliance Lead	Generate Annex IV technical documentation.

Tutorials

Writing Code-First Policy — Translate legal requirements into OSCAL controls.

Installation

pip install venturalitica

pip install "venturalitica[metrics]"

Adds Fairlearn for advanced fairness metrics.

pip install "venturalitica[dashboard]"

Adds Streamlit for the Glass Box dashboard.

pip install "venturalitica[agentic]"

Adds LangGraph, LangChain, and llama-cpp-python for agent workflows.

pip install "venturalitica[green]"

Adds CodeCarbon for energy consumption tracking.

pip install "venturalitica[full]"

Installs all optional packages (metrics + dashboard + agentic + green + torch).

Requires Python 3.11+.