Prove the log wasn't edited.
Paste one or more Agent Flight Recorder ledger records below (JSON, exactly as exported from /v1/actions). We recompute every SHA-256 hash server-side — the exact same code that wrote the chain — and return a verdict in seconds.
Try it — 30 seconds
Two buttons: load a demo ledger (real record format, labeled demo data), verify it, then tamper with it and watch the chain catch the edit.
Verification runs server-side with the identical hashing code that writes production ledgers. Pasted records are verified in memory and never stored. Rate-limited to keep the endpoint free for everyone.
Watch the chain catch an edit
Before: the record as written. After: the same record with one line of text changed. Try it yourself above — load the demo record, hit "Tamper the demo", then "Verify chain".
Before — as recorded
"seq": 2, "tool": "image_tool", "output": "Generated 1 image, 1024x1024 PNG (demo data)." hash: fd66eca4a8… ✓ matches
✓ VERIFIED — chain intact
After — one line edited
"seq": 2, "tool": "image_tool", "output": "Generated 1 image… [edited after writing]" recomputed hash ✗ does not match the stored fingerprint
✗ BROKEN at record #2
The edit changes record #2's fingerprint, so the link from record #3 to record #2 no longer lines up — the chain names the exact break point. Edits can't hide: every later record vouches for the ones before it.
What "tamper-evident" means, in plain language
Every record in an Agent Flight Recorder ledger contains a fingerprint (SHA-256 hash) computed from the previous record's fingerprint plus its own contents. It's like numbering the pages of a notebook and writing each page number in invisible ink on the next page: rip out page 50 and re-number the rest, and page 51's invisible ink no longer matches.
- VERIFIED = "chain intact — the log wasn't edited." Every fingerprint recomputed correctly, every link in order, from the first record to the last.
- BROKEN = the chain names the exact record where verification failed — either a record's contents were changed after writing, a record was removed, or the sequence was broken.
What this proves: that the records you hold are byte-identical to the records as written. What it does not prove: that the agent told the truth when it recorded the action — a ledger records what the agent reported, it can't audit the agent's honesty. That's why the verdict is about tampering, not truth.
Where else you can verify
- Dashboard: one-click "Verify chain" on every agent's page (logged in).
- API:
GET /v1/chain/{agent_id}returns chain status as JSON. - Badge:
/badge/{public_id}.svgembeds live chain status on your trust page. - Tamper-Seal certificates (Pro/Team): HMAC-signed attestations anyone can check at
/attest/<cert_id>— no account needed.
Frequently asked questions
Do I need an account to verify a record?
No. This page and the verification endpoint are public. Paste, verify, done.
What exactly should I paste?
The record objects returned by GET /v1/actions — or from a CSV/JSON evidence export. Each record needs seq, created_at, action_type, tool_name, input_summary, output_summary, metadata, prev_hash, and hash. Paste one record or many, in sequence order.
Does verification prove the agent behaved correctly?
No — it proves the record wasn't altered after it was written. The ledger is evidence of what the agent reported, which is exactly what auditors, customers, and courts need: an unedited account of events. Judging whether the actions were correct is the investigator's job.
Can someone edit a record and recompute all the hashes to hide it?
Only if they can also replace every independent copy of a later hash: exported CSVs, minted Tamper-Seal certificates, badge states, and any copy you hold. That's why verification works best against a record set you exported and kept — the attacker can't rewrite your copy.
Are my pasted records stored?
No. They're hashed in memory and discarded. Nothing you paste here touches a database.
How does this compare to AgentMon's approach?
AgentMon (Codenotary) is agent monitoring: it watches coding-agent sessions and flags cost, errors, and security findings — and its Start tier stores records in mutable local SQLite. It offers no public verification of agent records that we could find. Agent Flight Recorder is evidence: hash-chained, append-only records with public verification. Different jobs; see the honest comparison.
Want your own tamper-evident ledger? Record your agent's actions in five lines of code — free for 1 agent and 1,000 actions/month.