Proof of what an AI agent actually did.
AI agents write code, move money, and touch production — and nobody can independently verify what one actually did. korg is the open standard that fixes it: a tamper-evident ledger of agent actions, and a Certificate certificate anyone can re-verify offline, with zero trust in the tool that produced it.
Verify a Certificate
Drop a signed certificate of an agent session. The page re-derives the human summary from the events (so it can't lie) and checks the issuer's Ed25519 seal. The medallion stamps gold when valid — and cracks red at the exact tampered field.
Open the Certificate verifier → korg-ledger@v1 · raw sessionVerify a ledger
Drop a captured ledger.jsonl. The page recomputes the hash-chain, the causal
DAG, every per-event signature, and any anchors — entirely on your machine, using the same
verifier shipped in Rust, Python, and this JavaScript.
Replay a session
Drop a session or a Certificate and scrub the timeline. Watch the cumulative state — files touched, tools used — evolve step by step while the hash-chain verifies under every frame. A break lights up red at the exact step it happened.
Open the time-travel explorer →Why you can trust a green check
- You don't trust us. Verification recomputes everything from the bytes you hold — no server, no network, nothing phones home.
- Three independent implementations reproduce one frozen oracle, so a check in one is corroborated by two others written from the same spec.
- The summary can't lie: a Certificate's human-readable summary is re-derived from the events and rejected on any mismatch.