Settlement · Solana
On-chain programs
Three Anchor programs on devnet — agent-registry, mandate-registry, audit-anchor — hold agent identities, mandate commitments, and Merkle roots of audit batches.
Three layers with one job: decide before money moves, and leave evidence nobody has to take on faith. Chain characteristics below are properties of the underlying networks, dated and separate from Regent application performance.
Three Anchor programs on devnet — agent-registry, mandate-registry, audit-anchor — hold agent identities, mandate commitments, and Merkle roots of audit batches.
Full audit payloads as data-availability blobs with Merkle proofs. What is committed cannot be silently altered; what is stored where is documented, layer by layer.
Rule engine plus behavioural scoring (Isolation Forest with SHAP explanations, drift detection) — allow, hold for a human, or stop.
Dashboard for operators — agents, mandates, audit log, kill switch — and a Python SDK plus REST API for integration.
The agent is registered (AgentID), bound to a responsible party, and bounded by a mandate — amount, destination, time window, approval conditions. Nothing below runs without this.
Every financial action is authorized against the mandate and evaluated by Guardian. The deny happens before any money moves; the hold routes to a human operator.
The decision — allowed or denied — is hashed, Merkle-batched, and anchored. An auditor verifies against published keys and on-chain commitments, without trusting the operator.
Per-transaction anchoring at fractions of a cent is what makes per-action compliance economically viable. Cost and finality are network characteristics of Solana, not Regent performance promises — current figures live in the network's own documentation.
Dedicated DA for audit payloads (~$0.03–0.07 / MB, chain characteristic) with cryptographic guarantees on what was committed. Tamper-evidence is stated precisely: commitments are permanent; storage layers are documented per artifact.
Quarters, not promises: each item flips to “shipped” only when it is independently checkable.
Three Solana programs on devnet · Python SDK · REST API · Guardian behavioural scoring · operator dashboard with kill switch · live demo agents.
Mainnet beta · first design partners · published independent security review · bug bounty.
Compliance hardening: EU AI Act evidence mappings, SOC 2 groundwork, enhanced verification pipeline, institutional pilots.
Cross-chain audit anchoring via partner integrations; Solana remains the primary chain.
Language SDKs beyond Python are on the roadmap. The REST API works from any language today.