Capture what your AI did.
Verify the record.
Your agents take real actions on their own — approve refunds, move money, make calls for your customers. Swarrm gives each captured action a signed receipt in a tamper-evident log. If a decision is questioned, you hold a self-contained tamper-evident bundle whose integrity anyone can check offline, without an account.
“Our logs say so” isn’t proof.
Editable logs ask a later reviewer to trust whoever controlled them. A signed, checkpointed log gives that reviewer integrity checks they can rerun.
Your logs can be edited
Whoever controls a conventional database may be able to rewrite it. That weakens the record precisely when its history matters.
The AI vendor can’t vouch for itself
A platform grading its own AI’s homework is not enough. A separate verifier can test signatures and log structure without calling the producer.
Integrity checks have to survive the wait
Incidents and reviews show up months later. A self-contained bundle keeps the base integrity checks available to someone who was not there.
Four steps. No black box in the black box.
Plain enough to follow, real enough to hand your engineer or your auditor. The technical name for each step is on the right.
Recording is designed to stay outside the agent's critical path, and the gateway supports fail-open operation. Validate capture, persistence, and failure behaviour in your own setup before relying on those deployment-specific properties.
Evidence you can check yourself.
Each answer stays tied to the exact record and trust roots used.
Prove a captured receipt is intact
Every captured action is signed and placed in an authenticated log. Any change to the presented receipt or its authenticated path makes verification fail.
Make cyber tampering loud, not silent
Change the presented history and its checkpoint chain or Merkle proofs stop matching. Verification fails visibly.
Document and share the evidence trail
A bundle is self-contained. Whoever you share it with re-runs every base check offline, on their own machine, without asking Swarrm to vouch for it.
Confirm an agent comes from a trusted source
Counterparty Assurance checks an agent against an organisation root the relying party pins locally, plus current authority for the exact action. Favourable results remain non-authorizing until the external gates close.
Executive briefing
A one-page summary: what Swarrm proves today, how verification works, and what higher assurance requires.
Start free, on your own machine.
No account, no cloud bill, no sales call. Install it, run the demo, and inspect a locally verified tamper-evident record.
The questions everyone asks.
What is this, in one sentence?
Was every action captured?
Can I require durable capture for high-risk actions?
What if someone bypasses or stops the recorder?
ORPHAN — showing that something happened outside the recorded path, not what happened or who changed it.How does trusted-source confirmation work?
PASS_NOT_ENABLED with should_execute=false.