AI agents are making decisions on your behalf.
Nobody can prove what they actually did.
auditk measures the gap between what an AI agent declares it will do and what it actually does. The result is a calibrated drift score and a cryptographically signed evidence pack — tamper-evident, portable, and verifiable by any party with the public key.
Intent–Enactment Drift
Quantify the gap between what an AI agent said it would do and what it actually did.
Signed Evidence Packs
Ed25519-signed, canonical JSON artefacts. Self-describing, verifiable, portable.
Adapter Pattern
Works with Claude Code, OpenTelemetry, LangGraph, LangSmith, Langfuse, and raw traces.
Reference implementation · Cross-model benchmark · Apache-2.0
Follow the build on LinkedIn