fail-kit-fastapi 2.0.0

F.A.I.L. Kit - Forensic Audit of Intelligent Logic - FastAPI middleware

F.A.I.L. Kit - FastAPI Middleware

Zero-config FastAPI middleware for AI agent auditing.

Installation

pip install fail-kit

Quick Start

from fastapi import FastAPI
from fail_kit import fail_audit

app = FastAPI()

@app.post("/eval/run")
@fail_audit(auto_receipts=True)
async def evaluate(prompt: str, context: dict):
    # Your agent logic here
    result = await your_agent(prompt)
    return {
        "response": result["text"],
        "actions": result["actions"],
        "receipts": result["receipts"]
    }

Then run the audit:

pip install fail-kit-cli  # Or: npm install -g @fail-kit/cli
fail-audit scan
fail-audit run --format html

Usage

With Actions and Receipts

from fail_kit import fail_audit, FailAuditAction
from datetime import datetime

@app.post("/eval/run")
@fail_audit(auto_receipts=True)
async def evaluate(prompt: str, context: dict):
    # Create an action
    action = FailAuditAction(
        tool="database_query",
        input=prompt,
        output="Query result",
        status="success",
        latency_ms=45
    )
    
    return {
        "response": "Here's what I found...",
        "actions": [action.to_dict()],
    }
    # Receipts are auto-generated!

Simple Text Response

from fail_kit import fail_audit_simple

@app.post("/eval/run")
@fail_audit_simple
async def evaluate(prompt: str, context: dict):
    return f"Answer to: {prompt}"

With Escalation

from fail_kit import fail_audit

@app.post("/eval/run")
@fail_audit(auto_receipts=True)
async def evaluate(prompt: str, context: dict):
    if "delete" in prompt.lower():
        return {
            "response": "This requires human approval.",
            "actions": [],
            "policy": {
                "escalate": True,
                "reasons": ["high-stakes operation"]
            }
        }
    
    return {"response": "Done!", "actions": []}

API Reference

@fail_audit(auto_receipts=True, action_logger=None)

Decorator for agent endpoints.

• auto_receipts: Automatically generate receipts from actions
• action_logger: Optional async function to log actions

@fail_audit_simple

Simple decorator for text-only agents.

FailAuditAction

Helper class for creating actions:

action = FailAuditAction(
    tool="tool_name",
    input="input data",
    output="output data",
    status="success",  # or "failed"
    latency_ms=100
)

FailAuditReceipt

Helper class for creating receipts:

receipt = FailAuditReceipt(
    tool="tool_name",
    status="success",
    proof="Proof of execution"
)

Links

• Documentation
• Easy Integration Guide
• CLI Package
• Examples

License

MIT