Fiducia
RLUSD treasury copilot · XRPL · LATAM-first

Signed, verifiable treasury management for RLUSD startups.

Fiducia is an AI treasury copilot for XRPL. An autonomous agent manages your on-chain allocations, respects a deterministic policy you author, and produces a cryptographically verifiable audit trail of every decision.

fiducia — test:demo
$ pnpm test:demo

 Scenario A (expect: execute_after_retry)
  Tick 1 · proposal: 84 RLUSD → blocked (exceeds max_single_tx)
  Tick 2 · proposal: 42 RLUSD → pass
  tx: BF1ACEBFF388…AB2932

 Scenario B (expect: execute_after_approval)
  proposal: 65 RLUSD → require_approval → [AUTO-APPROVED]
  tx: 968882304…AB3C8

 Scenario C (expect: block)
  proposal: 540 RLUSD → blocked (3 independent rules)

✓ ALL 3 SCENARIOS PASSED

The gap we're filling

Treasury management is either too manual or too opaque. Babysitting a multisig doesn't scale past ten transactions a week. Parking everything in one vault violates concentration policy. Custom scripts don't produce logs an auditor can use. Institutional treasury tools start at $100M AUM.

A startup CFO managing $500K–$5M of RLUSD has nothing built for them — and their regulator, their board, and their reinsurer all want audit-grade records anyway.

The architecture of trust

Three primitives, in this order, every tick. The LLM is smart but not trusted. The YAML is dumb but authoritative. The ledger is cold but permanent. That's the compliance moat.

01
Claude proposes.

The planner reads your wallet state and market data, then proposes a single action with a natural-language rationale.

02
A YAML policy engine disposes.

A deterministic rulebook you author evaluates the proposal. Blocks anything unsafe — no matter how clever the LLM sounded.

03
The ledger remembers.

An Ed25519 signature of every decision is embedded in the XRPL transaction memo. Verifiable by any auditor with a one-command CLI.

Live demo — three scenarios, three signed records

Everything on this page is running code. These are real testnet transactions, each one carrying a cryptographic hash of the decision's signed provenance record.

A — policy-chunking story
Drift rebalance with retry loop

The agent proposes moving 84 RLUSD to rebalance. Policy blocks — exceeds the 75 single-tx limit. The agent re-proposes 42 RLUSD. Policy passes. Payment submits with the provenance hash in the memo.

tx · 42 FRL float → reserves (after 84 blocked, retried 42)
BF1ACEBFF3882172832C64B88766E1453A62135DD2A95F64FEF9F64E11AB2932
View on testnet.xrpl.org ↗
B — human-in-the-loop story
Claim event with explicit approval

A parametric oracle fires. Agent proposes a 65 RLUSD claims-wallet topup. Amount is above the approval threshold — policy returns require_approval. Operator types y. The approval outcome is signed as part of the provenance payload.

tx · 65 FRL float → claims (after human approval)
9688823042A8096BC771CE56AA5515755856BA1325443ECA6FF4AB70108AB3C8
View on testnet.xrpl.org ↗
C — policy-refuses-LLM story
Depeg alert, catastrophic proposal, blocked

Low-volume secondary venue shows RLUSD at 94¢. The LLM 'tries to help' and proposes rotating 540 RLUSD into XRP. Policy refuses on three independent grounds. No transaction executes. The refusal itself is the signed artifact.

No transaction executed.
Policy refused on three independent rules. The signed "blocked proposal" record IS the artifact.
The killshot
pnpm verify <tx-hash>

This is what an external auditor runs. They're not trusting us — they're verifying.

✓  tx found on ledger
✓  memo type    : fiducia/v1
✓  off-chain record found (scenario=A, action=rebalance)
✓  payload hash recomputed and matches on-chain memo
✓  Ed25519 signature valid for pubkey 76f053d22e70d903…

✓ AUDIT PASS — decision is cryptographically verified

Reproduce locally in under 3 minutes: pnpm setup:demo && pnpm setup:stablecoin && pnpm test:demo

How it fits together

We're T54-complementary, not T54-competitive — we consume their x402 facilitator layer rather than rebuild it. We route to existing yield vaults (Doppler, Soil, Kairo) rather than compete with them. The agent's decisions are our moat, not the execution primitives.

Planner (Claude)  →  Policy Engine (YAML)  →  Provenance Logger (Ed25519 + XRPL memo)
            │                                             │
            ▼                                             ▼
     x402 Client (T54)                   XRPL Executor  →  Doppler / Soil / native AMM / DEX
T54 Labs

Owns the x402 facilitator + Evernorth AI treasury partnership. We consume their rails.

Hummingbot / Renora

Trader and robo-advisor respectively. We're treasury-focused, audit-grade, XRPL-native.

Doppler / Soil / Kairo

Yield vaults. We route to them, not replace them.

Who it's for

Target segment: $100K–$10M RLUSD treasuries. Below Evernorth (T54's institutional partner). Above Renora (retail robo-advisor). Fit: Web3 startups with multi-wallet treasuries, regulated digital-asset funds, parametric insurers, DAO treasuries with audit requirements.

If a human is currently babysitting your multisig and your auditor has to trust spreadsheets — we're built for you.

LATAM-first, not LATAM-tagline

Most treasury tooling assumes a San Francisco operator. We don't. Built in Hermosillo, Sonora, México — we understand the regulatory landscape of Mexico's CNSF digital-asset pilot, Brazil's CVM framework, Colombia's emerging guidance.

LATAM crypto-native startups hit audit requirements earlier than their US counterparts, with fewer off-the-shelf tools. That's the wedge.

Who's building this

Luis Fernando Mendoza
Founder · Hermosillo, México

Software engineer, 10+ years. Trading & web3 focus. Applying to RippleX Ecosystem Programs for $150K over 6 months to ship M2 (x402 integration) and M3 (provenance dashboard), and onboard three LATAM design partners.

luisfernandomendozav@gmail.com

Looking for three LATAM design partners.

If you manage an RLUSD treasury between $100K and $10M — or know someone who does — I'd love a 20-minute call. No fees. Co-designed policy. Priority on the feature roadmap.