Decision trust protocol · API v0.7

Trust infrastructure for agent‑initiated payments.

AI agents will initiate trillions in transactions. Mandate Labs is the real-time decision verification layer between autonomous agents and the financial system — seven decision gates and two system steps, resolved in 23 milliseconds on existing rails.

Book a demo Read the docs

$ pip install mandate-labs-mcp — try the sandbox, no sign-up

23 msP50 decision

7 + 2Gates + system steps

99.95%Availability SLA

150K/hrValidated throughput

Mandate Labs back office — real-time decision trust overview with fleet trust score, authorization pipeline, and outcome split (demo data) — The back office · demo environment — curated sample data

See it work

One API call. Seven decision gates.

Send an authorization request and watch the Decision Trust Protocol evaluate it in real time — every gate, every score, every millisecond.

authorization_request.py

import mandate

response = mandate.authorizations.create(
    agent_id="agent_procurement_01",
    amount="2400.00",
    currency="USD",
    merchant="mch_aws_7291",
    intent_context={
        "intent_type": "PURCHASE",
        "task_reference": "TASK-2847",
        "reasoning_summary": "Scaling GPU
  instances for Q3 training run.
  Compared 3 providers on cost,
  latency. AWS optimal at $2,400.",
        "confidence": 0.94,
        "alternatives_considered": 3
    }
)

Authorization pipeline 0 ms

Agent Resolve1.2ms

Intent Verification3.8ms

Anomaly Gate (EDQS)7.2ms

Mandate Enforcement9.6ms

Behavioral Overlay12.4ms

Risk Scoring16.1ms

KYA Decision19.5ms

Persist + WebhooksSYSTEM21.8ms

Async Post-DecisionSYSTEMpost-resp.

EVALUATING… KYA 0.00 ZONE GREEN

Inside the trust layer

The console your compliance team will live in.

Every decision fully explainable — which gates fired, the KYA score at decision time, what the behavioral trust state was at the moment of decision, and the agent's attested intent. What you see is the product, in its shareable demo environment.

Decisions that need a human, ranked by risk.

Pulse is the triage surface: live fleet trust, the queue of held decisions, and a case file that explains in plain language why the protocol said no.

Case narrative: which gates failed, and why — no black boxes
Fleet trust live, with every spike annotated
From flag to disposition without leaving the screen

Pulse triage — held decisions ranked by risk with a full case file — dashboard.mandatelabs.ai — demo environment, curated sample data

The agent suite

Operators that make your compliance team faster.

A brain that knows the protocol cold, and an investigator grounded in your case data. Assistive by design — humans keep every decision.

Ivy

Knowledge copilot · read-only

Always-on expert on the platform, the DTP, scoring, and the dashboard. Teaches, explains, and walks investigations — with no access to your data, by design.

Eva

AML investigator · recommend-only

PAID ADD-ON

Reads the case's real data: triages alerts, correlates look-alike activity, and drafts regulator-ready SAR/STR narratives. A human dispositions every alert and files every report.

Both run inside your tenancy, entitlement, and audit model — one registry, built for more operators over time.

Case investigation with Eva's structured assessment: key signals, recommended disposition, SAR-worthiness — Eva investigation report — demo environment, curated sample data

Portable trust

A credit report for your agent.

The Agent Trust Report is a certified, signed KYA profile — score composition, EDQS weighting, trajectory, and history — that an agent can present anywhere.

Two-page certified document: the report plus a Certificate of Authenticity — report ID, SHA-256 content digest, Ed25519 signature, QR verification
Anyone can verify against Mandate's published keys (JWKS) — offline, no callback
Score composition across identity, mandate coverage, behavior, and principal standing
Valid 90 days from issuance — trust scores drift as agents act

Request a sample report

Agent trust report — KYA score 94/100 with score composition and trust trajectory — Agent trust report — demo environment, curated sample data

The protocol

Seven decision gates. Two system steps.

Every authorization traverses the full Decision Trust Protocol pipeline — from agent resolution to post-decision analysis. Decisions: APPROVE, DECLINE, or STEP_UP.

01Agent resolveIdentify the agent; load its mandate, trust tier, and behavioral profile from the registry.

02Intent verificationParse the intent context; validate reasoning quality, confidence calibration, and task alignment.

03Anomaly gateSix-dimensional anomaly detection against the agent's behavioral baseline, scored with EDQS.

04Mandate enforcementVerify the transaction against spending limits, merchant allowlists, and time-of-day policies.

05Behavioral overlayAssess the agent's current behavioral state — degradation, drift, and prompt injection — before the network sees it.

06Risk scoringComposite risk across merchant reputation, transaction patterns, velocity, and geographic signals.

07KYA decisionGenerate the Know Your Agent trust score and map it to a trust zone.

S1Persist + webhooksWrite the decision to the hash-chained audit log; fire webhooks for downstream systems.

S2Async post-decisionEnqueue behavioral model updates, regulatory checks, and cross-agent correlation.

GREEN ≥ 0.70 AMBER 0.40–0.70 RED 0.20–0.40 CRITICAL < 0.20

Settlement

Two rails. One trust engine.

The same KYA scoring, mandate enforcement, and intent verification govern both rails — the rail only selects the transaction context and the form of evidence.

Card — default

Card-issuing rail

Mastercard today. Decisions map onto card-scheme data elements — structured risk assessment designed for DE 48.75 — with clearing and settlement flowing back as outcomes that close the trust loop.

MCC · merchant · country · velocity constraints
DE 48.75-mapped advisory payloads
Scheme settlement → CLEARED / SETTLED outcomes

Crypto — opt-in

x402 stablecoin rail

Base L2, USDC. On approval, a signed attestation (EdDSA JWS) is bound to the exact EIP-3009 transfer — the wallet keeps custody and verifies independently. A settlement watcher reconciles on-chain and records the outcome.

x402 pay-per-call · USDC on Base (8453)
Transfer-bound attestations — no replay, no substitution
Settlement watcher → SETTLED on confirmation

Verifiable Intent v0.1 aligned ISO 8583 / ISO 20022 Mastercard DE 48.75 3-D Secure 2.0 SOC 2 Type II in progress From the team behind PayCaddy

Research

Advancing the science of agent trust.

Every scoring model, behavioral metric, and trust threshold is grounded in published research. We publish openly — the frameworks need to be scrutinized to become standards.

Jun 2026 Economic Decision Quality Score (EDQS)A six-dimension composite metric for evaluating AI agent economic reasoning, independent of outcome. Read → Jun 2026 Agent Behavioral TelemetryBehavioral drift as a leading indicator of agent compromise — the seven-signal fingerprint. Read → May 2026 The Decision Trust ProtocolA layered authorization framework for autonomous agent commerce on existing rails. Read →

Deployment

Two integration modes. One protocol.

Both return the same enriched, advisory decision — the difference is when DTP is consulted. Your engine keeps final authority.

Issuer-side Inside the authorization windowYour processor consults DTP during decisioning; KYA score, risk signals, and reason codes return inside the window. Production is sales-led, certified via Mandate Certified Integration. Integration guide → Platform-side Before the transaction existsYour agent platform calls DTP pre-presentment — only approved intents become transactions. Sandbox is self-serve today: pip install mandate-labs-mcp. Quickstart →

The infrastructure for agent commerce is being built now.

Join the institutions shaping the trust layer for autonomous transactions.