Cryptographic audit trails
for every AI output.
A lightweight Python package that wraps your existing AI API calls and creates independently verifiable records. Prove exactly what your AI produced, when, and from what input — with cryptographic guarantees.
Three lines to add provenance
# Before: no audit trail
from openai import OpenAI
client = OpenAI()
# After: every call is recorded
from provyn import wrap
client = wrap(OpenAI())
# Use exactly as before — zero code changes
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "..."}]
)
# Every call now has a verifiable record:
# - SHA-256 hash of input
# - SHA-256 hash of output
# - Model version, timestamp, latency
# - Chained to previous recordWhat every record contains
Input Hash
SHA-256 hash of the full prompt/input payload. Proves exactly what was sent to the model.
Output Hash
SHA-256 hash of the model's complete response. Proves the output hasn't been modified.
Model & Version
Exact model identifier and version string. Tracks which model produced which output.
Timestamp
ISO 8601 timestamp with timezone. Establishes when the AI interaction occurred.
Chain Hash
SHA-256 hash linking this record to the previous one. Tamper-evident sequential chain.
Human Review
Optional flag indicating whether a human reviewed the output before use.
Who needs this
Compliance Officers
Demonstrate to auditors that your AI systems have traceable, tamper-proof records. Required under the EU AI Act for high-risk systems.
Insurance Underwriters
Assess whether AI-using applicants have provenance controls in place. Price risk based on auditability — not just deployment.
Legal Counsel
Produce verifiable evidence of what an AI system generated and when. Critical for liability disputes, discovery, and regulatory defense.
Engineering Leaders
Add audit infrastructure without changing your AI integration code. Drop-in wrapper for OpenAI, Anthropic, and Google APIs.
Architecture
Technical Specification
Supported Providers
Record Format
{
"record_id": "uuid-v4",
"chain_hash": "sha256:...",
"input_hash": "sha256:...",
"output_hash": "sha256:...",
"model": "gpt-4o-2024-08-06",
"provider": "openai",
"timestamp": "ISO-8601",
"latency_ms": 1423,
"human_reviewed": false,
"prev_record": "sha256:..."
}Deployment Options
Verification API
Pricing
Free
- ✓1,000 records/month
- ✓Single project
- ✓7-day retention
- ✓Community support
Pro
- ✓50,000 records/month
- ✓Unlimited projects
- ✓1-year retention
- ✓Verification API
- ✓Priority support
Enterprise
- ✓Unlimited records
- ✓Self-hosted option
- ✓Unlimited retention
- ✓SSO & RBAC
- ✓SLA & dedicated support
- ✓Custom integrations