model-witness 0.4.0

Cryptographic audit infrastructure for AI inference

model-witness

Model Witness provides cryptographic audit infrastructure for AI inference. Every prompt and response is hashed, signed, and anchored to an immutable ledger — giving you a tamper-evident chain of evidence for every AI call your application makes.

This package is currently in pre-alpha. Full SDK documentation and installation instructions will be published with the v0.1.0 release.

For early access enquiries contact [e-mail].

---

Accessing the record ID

Every inference call produces a record_id you can store alongside your own application data to link back to the audit trail.

Pattern A — last_record_id() (simple, single-threaded)

from model_witness import ModelWitness
import openai

client = ModelWitness(openai.OpenAI())

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": user_input}]
)

record_id = client.last_record_id()
db.save({"application_id": app_id, "mw_record_id": record_id})

The record_id is set synchronously before create() returns — there is no added latency to the LLM call.

Warning: last_record_id() is not thread-safe for concurrent inference calls. If you are making multiple simultaneous LLM calls, use the on_record callback instead — each invocation receives its own record with the correct record_id isolated to that call.

Pattern B — on_record callback (concurrent usage)

from model_witness import ModelWitness
import openai

def handle_record(record):
    db.save({"application_id": app_id, "mw_record_id": record.record_id})

client = ModelWitness(
    openai.OpenAI(),
    on_record=handle_record,
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": user_input}]
)

The on_record callback is called synchronously on the same thread as the LLM response, before the record is shipped to the server. It receives the full InferenceRecord object including record_id, input_hash, output_hash, and token counts.