Introduction

Evidence infrastructure for AI decisions. Cryptographic proof of human oversight for EU AI Act, LGPD, SOC 2.

What XASE Does

Capability	Description
Decision Capture	Record every AI decision with input, output, and model metadata
Human-in-the-Loop	Immutable proof of human review, approval, or override
Explainability	Automatic SHAP/LIME explanations stored with each decision
Model Registry	Track which model version made each decision
Evidence Export	Generate offline-verifiable bundles for audits

What XASE is NOT

✗A workflow tool (we record actions, not manage them)

✗A model training platform (we don't touch your ML)

✗A monitoring tool (we don't track model drift)

✗Blockchain (we use hash chains, much simpler)

How It Works

Your AI System          XASE                Auditor
   │                   │                     │
   ├── Decision made ──────────►             │
   │                   │ Record + Hash       │
   │                   │ + Sign              │
   │                   │                     │
   ├── Human reviews ──────────►             │
   │                   │ HITL + Identity     │
   │                   │ + Timestamp         │
   │                   │                     │
   │                   │ ◄────── Request audit
   │                   │                     │
   │                   ├── Export Bundle ───────►
   │                   │                     │
   │                   │         [Verify offline: ✓ Valid]

XASE bundles are verifiable offline. Auditors don't need to call our API or trust our infrastructure.

Quick Example

example.pypython

import xase

client = xase.Client(api_key="xase_pk_...")

# 1. Record AI decision
record = client.records.create(
    model_id="credit-model-v1",
    input={"customer_id": "cust_123", "income": 85000},
    output={"decision": "APPROVED", "limit": 25000}
)

# 2. Record human intervention
intervention = client.records.intervene(
    record_id=record.id,
    actor_email="analyst@company.com",
    action="APPROVED",
    reason="Documentation verified"
)

# 3. Export for audit
bundle = client.exports.create(record_id=record.id)
bundle.download("./evidence.zip")

Next Steps

Quickstart

Get running in 5 minutes

Core Concepts

Understand the architecture

Python SDK

Full installation guide

Was this helpful?

Edit on GitHub