SolGuard is a pre-signing security gateway that constrains autonomous-agent payments before a wallet authorization can reach settlement.
The initial buyer hypothesis is an agent platform or wallet provider that allows software agents to purchase APIs, data, or services. Its risk is not merely that a payment fails: a compromised or misdirected agent can issue valid-looking payments at machine speed while still holding legitimate wallet access.
An owner delegates a small financial mandate to an agent. SolGuard evaluates each canonical payment request against:
- per-agent maximum spend and recipient allow/block rules;
- amount, velocity, recipient-novelty, and compound-drain detection;
- request expiry and per-agent nonce replay protection;
- a fail-closed decision gateway; and
- a short-lived, exact-request authorization consumed once at the wallet boundary.
The result is ALLOW, REQUIRE_APPROVAL, or BLOCK, with stable reason codes and a
sanitized audit receipt. A block produces no signing authorization.
The prototype enforces the decision before wallet settlement, rather than alerting after funds move. The demonstrated build also keeps the security engine independent of the payment rail: the same controls drive an in-memory fallback and one optional Pay.sh sandbox adapter.
This is a verified prototype property, not a claim that SolGuard supports every wallet, protocol, or production threat model.
A possible entry model is usage-based screening for agent platforms and wallets, followed by paid policy management, audit retention, and enterprise controls. No customer, revenue, pricing-validation, or market-share claim is made.
We are seeking one agent-platform or wallet team to validate:
- where a pre-signing decision can be enforced in its real transaction path;
- which mandate controls its users can understand and operate safely;
- what latency and availability budgets that boundary requires; and
- which audit evidence its security and compliance teams need.
The next validation step is a scoped sandbox integration using the partner's own payment intents and adversarial cases, with no production funds.
The current build is not production-ready. It uses process-local state and does not provide durable replay coordination, multi-tenant isolation, authenticated policy administration, hardened key custody, high availability, incident response, or an independent security review. Pay.sh activity is sandbox-only; the local balance and settlement path are simulated. The x402 v2 adapter currently proves Solana-devnet request mapping and pre-signing ordering only; its settlement remains simulated.
SolGuard does not custody keys, replace wallet cryptography, guarantee fraud prevention, or claim that behavioural rules can identify every attack.