Skip to content
GARD

Developers

Add GARD before execution.

Give autonomous applications a policy and transaction-decision layer with a simple integration.

Integration

Live on testnet
agent.ts
import { Gard } from "@gard/sdk";

const gard = new Gard({ apiKey: process.env.GARD_API_KEY });

const result = await gard.evaluate({
  recipient,
  asset: "USDC",
  amount: "500.00",
  purpose: "Purchase GPU compute",
});

if (result.decision === "ALLOW") {
  await wallet.send(tx);
}

API

POST /v1/evaluate

Submit a proposed transaction and receive a decision before anything is executed.

Live on testnet
Request
POST /v1/evaluate
Authorization: Bearer $GARD_API_KEY

{
  "recipient": "0x3e88...a6f7",
  "asset": "USDC",
  "amount": "500.00",
  "purpose": "Purchase GPU compute"
}
Response
{
  "version": "1",
  "decision": "REVIEW",
  "confidenceBps": 9000,
  "reasons": [
    { "code": "new_recipient",
      "message": "Recipient has no successful history" }
  ],
  "evaluator": { "provider": "gard-rules" },
  "accessMode": "api_key",
  "pricingTier": "REGISTERED"
}

Live on Arc Testnet with free test assets. Mainnet is not live. Full reference: /v1/openapi.json.

Integration lifecycle

Six steps from request to record.

  1. 1

    Create request

    Your agent or app builds the proposed transaction with its purpose.

  2. 2

    Submit to GARD

    Send the request to the evaluate endpoint before signing.

  3. 3

    Receive decision

    GARD returns ALLOW, REVIEW, or BLOCK, with reasons.

  4. 4

    Apply application policy

    Your application decides how to handle each outcome.

  5. 5

    Execute or stop

    Proceed on ALLOW, route to a person on REVIEW, halt on BLOCK.

  6. 6

    Record result

    Store the decision and outcome for audit and analysis.

Developer benefits

Built for the people wiring up autonomous money.

  • Simple API

    One request, one decision. Minimal surface to integrate.

  • Structured output

    Machine-readable decisions and reason codes.

  • Policy-first architecture

    Explicit rules sit alongside risk evaluation.

  • Auditability

    Decisions are designed to be recorded and reviewable.

  • Designed for autonomous systems

    Built for software that spends without a person in the loop.

SDKs and interfaces

SDKs and MCP.

Built and tested in the repository; not yet published to a package registry.

  • Built

    TypeScript SDK

    Typed client with idempotent retries and webhook verification.

  • Built

    MCP server

    Five tools so an AI agent checks every payment first.

  • Live on testnet

    REST API

    Plain HTTPS and JSON for any language, with an OpenAPI spec.

  • Live on testnet

    Webhooks

    Signed, retried notifications for decisions and executions.

See the flow end to end.

Read the documentation, or try the demo with a mock evaluator.