DecisionTrace

DecisionTrace is a simple, powerful tool for creating an audit trail of your AI's decisions. Think of it as a flight data recorder for your AI.

It ensures that every important decision your AI makes is logged in a secure, tamper-evident way, so you can always go back and understand why a decision was made.

Features

• Append-Only Log: Creates a secure, unchangeable log of AI decisions.
• Integrity Chain: Uses a hash chain (like a lightweight blockchain) to prevent tampering and guarantee the chronological order of decisions.
• Replay Capability: Easily "replay" any past decision to see the full context: the prompt, the model used, the output, and more.
• Verification: A simple command to verify the entire log's integrity and detect any corruption or tampering.

Quick Start

1. Installation

First, clone the repository. Then, navigate into the decisiontrace directory and install the tool:

# Install the package in editable mode
pip install -e .

This adds the decisiontrace command to your system.

2. How to Use

Log a Decision

Create a .json file (e.g., my_decision.json) with your AI's decision details:

{
  "model": "llama3.2:1b-instruct",
  "config": { "temperature": 0.7 },
  "prompt": "Should we approve this loan?",
  "context_sources": ["policy_v4.pdf", "user_credit_report.json"],
  "output": "Approval recommended based on high credit score.",
  "confidence": 0.95,
  "risk_flags": ["user_pii_involved"]
}

Then, log it:

decisiontrace log my_decision.json

Replay a Decision

Use the decision_id from the log output to see the full details of a past decision:

decisiontrace replay <your-decision-id>

Verify the Log

Check the entire history for tampering:

decisiontrace verify

That's it! You now have a secure, auditable log of your AI's decisions.