A log tells you what happened, according to whoever kept the log.
An agent sends a payment, calls a tool, or exports a customer record. The only trace it leaves is a log, and whoever runs that log controls it.
A log can be edited, rotated, or quietly dropped. It can’t show who approved an action or rebuild what actually ran. So when an auditor or an incident asks for proof, all you have is a claim.
09:41:52 payments.transfer payee="Clark Kent"
09:41:52 amount=$42,000.00$4,200.00
09:41:53 status=ok · approved_by=policy
the same edit, against both records · $4,200.00 → $42,000.00
You need proof, not a nicer log.
A log is only as honest as whoever keeps it — so HESO hands you proof instead: every action decided against your policy, co-signed by a person when it matters, and verifiable by anyone, offline.
Every action your agent takes (a payment, a tool call, an HTTP request, a data export) is caught before it runs and checked against your policy. The first rule that matches decides what happens: allow it, block it, redact a sensitive field, or hold it for a person. Nothing executes until it clears.
High-stakes actions stop and wait for a human. They approve right in the console, and their browser co-signs with a key that lives on their device. The cloud never sees that key, so no one can forge the sign-off, including us. Need more than one yes? Set a quorum, like 2 of 3 approvers, and each person signs only their own decision. The policy floor you set can’t be bypassed by anyone.
The result is a receipt: the exact bytes of the action, signed with Ed25519 and chained to the one before it with BLAKE3. Drop it into any browser and the check runs right there on the page. No network, no account, and no need to trust HESO.
A receipt is evidence you hold, not a status we report. The check runs in your browser, not on our servers and not on our word. Try it: edit the amount below.
The recomputed hash matches the signed hash, so the signature holds. This receipt is authentic, verified offline in your browser.
to decide: allow, block, redact, or ask a human
signature on the exact bytes of every action
chain that links each receipt to the last
offline verification that needs nothing from us
A receipt also exports to open formats like AERF and COSE, so other tools can read the evidence. It isn’t a private lock-in format.
The EU’s logging rules for high-risk AI land in December 2027, and a hash-chained history can’t be backfilled after the fact. The audit trail has to already exist when the deadline arrives.
Paid tiers buy cloud scale, retention, and team features. The proof itself is on every plan.
For individuals and small experiments
$0forever
hard cap · no surprise bills
7 days retention · 5 req/sec
For growing teams shipping AI to production
$29/ mo
then $1 per 1,000 receipts past the floor
30 days retention · 20 req/sec
For teams that need scale and longer retention.
$299/ mo
then $0.50 per 1,000 receipts past the floor
6 months retention · 60 req/sec
Unlimited scale with dedicated support and SLAs
Contact
uncapped cloud mirror
1 year retention · 150+ req/sec
Only the cloud mirror is metered. You’re billed past the floor, and never blocked until the 10× runaway ceiling. The local signed chain always runs: no account, no network, never refused. Free has no overage. The mirror just stops.
Embed the gate in your own product and a customer co-signs the actions that touch them, with the same device-held keys and the same signed receipt. It’s included on every plan with a small “powered by heso” mark. On Team it wears your colors, with no mark.
Let your customers approveRun the March campaign for $1,250?
Your agency’s agent wants to act on this account. Your call.
A governed agent is one signing identity and policy scope, not a concurrency limit.
Open a receipt and check it in your browser. Ed25519 and BLAKE3, verified on your own machine with nothing from HESO.
free forever · 1 agent · no card
