Skip to content
GARD

Product

Transaction protection before execution.

Most wallets and agents know how to execute a transaction. GARD is being built to help them decide whether they should.

Transaction request

Illustrative
Wallet
0x9c41…1d73
Recipient
0x7a3f…d9c4
Asset
USDC
Amount
500.00
Purpose
Purchase API credits

GARD decision

ALLOW
  • Intent matches purpose
  • Within policy limits
  • Known recipient

Transaction evaluation

Context in. A clear decision out.

GARD is designed to take the full context of a proposed transaction and return one of three decisions your application can act on.

Input

  • Wallet
  • Recipient
  • Asset
  • Amount
  • Intended purpose
  • Policy
  • Transaction history and context

Output

  • ALLOWWithin policy. Safe to proceed.
  • REVIEWNeeds a human or additional checks.
  • BLOCKViolates policy or looks high risk.

Policy engine

Rules your application defines.

Policies make spending behavior explicit. They are intended to be evaluated on every request, alongside risk signals.

Example policy

Illustrative
policy.json
{
  "maxTransaction": 5000,
  "dailyLimit": 10000,
  "allowedAssets": ["USDC"],
  "unknownRecipient": "review",
  "unknownContract": "review",
  "humanReviewThreshold": 1000
}
  • Maximum transaction value

    Cap the size of any single transaction.

  • Daily spend limits

    Bound total spend over a rolling period.

  • Approved assets

    Allow only the assets an agent should use.

  • Approved recipients

    Maintain allowlists of trusted destinations.

  • Unknown-recipient review

    Route first-time recipients to review.

  • Contract interaction restrictions

    Restrict or review calls to unverified contracts.

  • Human-review thresholds

    Require a person above a set amount.

Risk intelligence

Signals that inform every decision.

These are the intended signal categories for GARD's risk evaluation. They are part of the design and are not yet live.

Planned
  • Transaction context

    Purpose, timing, and the surrounding activity of a request.

  • Recipient history

    Prior interactions between the wallet and the recipient.

  • Contract characteristics

    Properties of the contract a transaction touches.

  • Behavior deviation

    How a request differs from an agent's normal pattern.

  • Amount anomalies

    Sizes that are unusual for the wallet or purpose.

  • Merchant and recipient trust context

    What is known about the counterparty.

Decision history

An audit trail for autonomous spending.

Each decision is intended to be recorded with its inputs and reasons, so teams can review what an agent tried to do and why GARD responded as it did.

  • dec_8f2a91

    Send 500 USDC

    ALLOW

    Illustrative, 10:42

    within_policyknown_recipient
  • dec_8f2a92

    Send 1,800 USDC

    REVIEW

    Illustrative, 10:47

    new_recipienthuman_review_threshold
  • dec_8f2a93

    Send 9,200 USDC

    BLOCK

    Illustrative, 11:03

    exceeds_max_transaction

Illustrative records. Not real transactions.

Integration architecture

Sits between the agent and execution.

GARD is designed to be a single decision step in front of the wallet or execution layer. Nothing moves until the application receives a decision and applies it.

  1. Agent / App

    Proposes a transaction

  2. GARD

    Evaluates intent, risk, and policy

  3. Decision

    ALLOW, REVIEW, or BLOCK

    ALLOWREVIEWBLOCK
  4. Wallet / Execution

    Executes only when permitted

Protected by GARD

A label that says what it means — and what it does not.

Applications that integrate GARD may display a “Protected by GARD” badge. It indicates that a transaction passed through GARD's policy and evaluation workflow.

Protected by GARD

It means

The transaction was evaluated against intent, risk, and the application's policy before execution.

It does not mean

  • Insured
  • Guaranteed safe
  • Guaranteed reversible
  • Free from financial risk

GARD provides policy and risk evaluation. A GARD decision is not insurance or a guarantee against loss.

Start building with GARD.

Put a decision in front of every autonomous transaction.