TypeScript SDK

@verglas/sdk (in-repo package, sdk/) — viem-based reads and writes for every Verglas surface, plus Node-only proving helpers.

Install & wire

import { VerglasClient, FUJI_DEPLOYMENT, TREASURER_DEPLOYMENT } from "@verglas/sdk";

const client = VerglasClient.fuji(); // read-only against the live deployment

The one call that matters

// On any L1 with a VerglasGate — is this agent's trust fresh and sufficient?
const ok = await client.isCleared(FUJI_DEPLOYMENT.agentId); // → true

Reads

const state = await client.getAccountState();   // owner, agent, limits, commitment, frozen, whitelist
const att   = await client.getAttestation(219n); // latest hub attestation (requestHash, commitment, txCount, score, issuedAt)
const val   = await client.getValidationStatus(requestHash); // canonical 8004 registry stamp

Writes (need a wallet client)

await client.spend(to, amount);                       // as the vault's agent
await client.submitProof(agentId, requestHash, ...);  // prove a window
await client.carryToDispatch(agentId);                // ICM carry to the live gate

Proving (Node only)

Deliberately split so browser bundles stay clean:

import { proveWindow, toCalldata } from "@verglas/sdk/prove";

const { proof, publicSignals } = await proveWindow({ spends, whitelist, perTxLimit, initialCommitment });
const calldata = toCalldata(proof, publicSignals); // ready for submitProof

Requires the build/ circuit artifacts (wasm + zkey). Window shape: N=64 spends, WL=8 whitelist slots — mirrors the circuit 1:1.

Constants

FUJI_DEPLOYMENT (hub, verifier, account, usdc, gateOnDispatch, agentId, deployBlock), TREASURER_DEPLOYMENT (treasurer, account, agentId, pyth, usdTryPriceId), BLOCKCHAIN_IDS, TELEPORTER_ADDRESS, IDENTITY_REGISTRY_ADDRESS — all checksummed, all covered by tests (npm test; add VERGLAS_LIVE=1 for a live smoke against Fuji).