Facto Python SDK

Forensic Accountability Infrastructure for AI Agents

facto-ai is the Python SDK for Facto, a system designed to create tamper-proof, verifiable audit trails for AI agent actions. It allows you to wrap LLM calls and agent steps, automatically generating cryptographically signed events that are stored in a verifiable log.

Features

• Cryptographic Signatures: Every event is signed with Ed25519, ensuring authenticity.
• Tamper-Proof Logging: Events form a hash chain, making the history immutable.
• Easy Integration: Use decorators or context managers to wrap your existing code.
• Verification CLI: Included CLI tool to verify evidence bundles offline.

Installation

pip install facto-ai

Quick Start

1. Initialize the Client

from facto import FactoClient, FactoConfig

client = FactoClient(
    FactoConfig(
        endpoint="http://localhost:8080",  # Your Facto ingestion endpoint
        agent_id="my-agent-001",
    )
)

2. Wrap LLM Calls

Using the Decorator:

from facto import ExecutionMeta

@client.factod(
    action_type="llm_call",
    execution_meta=ExecutionMeta(model_id="gpt-4")
)
def chat_with_llm(prompt):
    # Your LLM logic here
    return {"response": "Hello world!"}

result = chat_with_llm("Hi there")

Using the Context Manager:

with client.facto(
    "tool_execution",
    input_data={"tool": "search", "query": "weather in sf"}
) as ctx:
    # Perform action
    result = perform_search("weather in sf")
    
    # Record output
    ctx.output = result
    ctx.status = "success"

3. Verify Events

You can verify exported evidence bundles using the CLI:

facto verify path/to/evidence.json

Documentation

For full documentation, visit docs.facto.ai or the GitHub Repository.