iron-thread 0.1.0

Ruthless middleware that stops broken AI outputs from reaching your database

iron-thread

Ruthless middleware that stops broken AI outputs from reaching your database.

Install

pip install iron-thread

Quick Start

from ironthread import IronThread

it = IronThread()

# Create a schema
schema = it.create_schema(
    name="User Profile",
    schema_definition={
        "required": ["name", "email", "age"],
        "properties": {
            "name": {"type": "string"},
            "email": {"type": "string"},
            "age": {"type": "integer"}
        }
    }
)

# Validate any AI output
result = it.validate(
    ai_output='{"name": "John", "email": "[e-mail]", "age": 28}',
    schema_id=schema["id"]
)

if result.passed:
    print("Clean — safe to send to database")
    print(result.data)
else:
    print(f"Blocked — {result.reason}")

Real World Usage

import openai
from ironthread import IronThread

it = IronThread()
client = openai.OpenAI()

# Your AI call
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Generate a user profile as JSON"}]
)

ai_output = response.choices[0].message.content

# Validate before touching your database
result = it.validate(
    ai_output=ai_output,
    schema_id="your-schema-id",
    model_used="gpt-4"
)

if result.passed:
    db.save(result.data)  # safe
else:
    print(f"AI output rejected: {result.reason}")

API

IronThread(host=...)

Initialize the client. Defaults to the hosted API.

.create_schema(name, schema_definition, description="")

Create a validation schema. Returns the schema with its id.

.validate(ai_output, schema_id, model_used="unknown")

Validate AI output against a schema. Returns a ValidationResult.

.stats()

Get dashboard stats — total runs, pass rate, avg latency.

.runs()

Get the last 50 validation runs.

ValidationResult

result.passed      # True or False
result.status      # "passed" / "failed" / "corrected"
result.reason      # why it failed (if it did)
result.data        # the clean validated output
result.latency_ms  # how long validation took
result.run_id      # ID of this run in the database

Links

• API Docs: https://iron-thread-production.up.railway.app/docs
• GitHub: https://github.com/eugene001dayne/iron-thread