![Arc[Context]](/_next/image?url=%2Farc-context-logo.png&w=640&q=75)
A verifiable memory layer for AI agents
Agents deciding to move money shouldn’t trust the wording of a request. Arc[Context] gives every action cryptographic, independently-verifiable proof of who actually authored it — not an operator’s claim.
How it works
Proof, not promises
Every high-value action moves through the same four steps — deterministically, the same way every time.
Write the request
A request — a transfer, an approval — is written to Arkiv as a signed entity. Its native $creator is set by the network itself, from whichever wallet actually sent it. Nothing here is text a caller can spoof.
Verify provenance
The Verifier Agent reads the entity back and checks its $creator against an approved-wallet allowlist — the Trust Gate. This is the only question that matters: not what the request says, but who cryptographically stands behind it.
Execute or reject — fail-closed
APPROVED executes the action and logs it. REJECTED — or any error, timeout, or uncertainty — blocks it. There is no fallback to unverified execution; when in doubt, the system always rejects.
An operational record, not a claim
Every decision lands in an append-only audit trail: creator, verdict, and transaction hash. Anyone can verify it independently against Arkiv — not just trust what this dashboard shows.
Live demo
Watch it verify a request
Dispatch a pending action from the inbox, or submit your own, and follow it through Context → Action → Verification → Result in real time.
Agent Inbox
Pending actions, ready to dispatch. Press a number key or click Run.
Legitimate request
An approved wallet submits a transfer.
Expect: APPROVED → Executed
Spoofed request
An unapproved wallet submits the same transfer.
Expect: REJECTED → Blocked
Operational record
Jump to the full history of everything run in this session.
View the record
Operational Record
Every action run in this session — searchable, filterable, exportable.
No actions yet this session.