Skip to main content

​
Policy Engine

The Policy Engine evaluates declarative policies before every agent action. It enforces default deny — if no policy explicitly allows an action, it’s rejected.

​
Design Principles

  • Default deny — new agents can’t do anything until policies are assigned
  • Declarative — JSON-based policy definitions, no custom DSL to learn
  • Composable — multiple policies evaluated in order; ALL must pass
  • Fast — < 5ms evaluation, cached in Redis

​
Policy Types

​
Spending Limit

Controls how much value an agent can transact: 
{
  "id": "spending-limit-1",
  "type": "spending-limit",
  "enabled": true,
  "config": {
    "maxPerTransaction": "100000000000000000",
    "maxPerDay": "500000000000000000",
    "maxPerWeek": "2000000000000000000"
  }
}
Values are in wei (for EVM) or lamports (for Solana). The engine tracks cumulative spend per agent over rolling time windows.

​
Approved Addresses

Whitelist of addresses an agent can interact with: 
{
  "id": "approved-addrs-1",
  "type": "approved-addresses",
  "enabled": true,
  "config": {
    "addresses": [
      "0x1111111254EEB25477B68fb85Ed929f73A960582",
      "0xDef1C0ded9bec7F1a1670819833240f027b25EfF"
    ]
  }
}
Approved addresses work as a whitelist. If enabled, only transactions to listed addresses are allowed. Disable this policy to allow transactions to any address.

​
Auto-Approve Threshold

Transactions below this value are auto-approved; above requires manual approval: 
{
  "id": "auto-approve-1",
  "type": "auto-approve-threshold",
  "enabled": true,
  "config": {
    "maxValue": "50000000000000000"
  }
}
When a transaction exceeds the threshold but passes other policies, it enters the approval queue. A tenant admin must explicitly approve or reject it.

​
Rate Limit

Controls how many transactions an agent can submit: 
{
  "id": "rate-limit-1",
  "type": "rate-limit",
  "enabled": true,
  "config": {
    "maxPerMinute": 10,
    "maxPerHour": 100
  }
}

​
Time Window

Restricts when an agent can transact: 
{
  "id": "time-window-1",
  "type": "time-window",
  "enabled": true,
  "config": {
    "allowedHours": { "start": 8, "end": 22 },
    "timezone": "UTC"
  }
}

​
Allowed Chains

Restricts which blockchain networks an agent can use: 
{
  "id": "allowed-chains-1",
  "type": "allowed-chains",
  "enabled": true,
  "config": {
    "chainIds": [8453, 1]
  }
}

​
Evaluation Flow

Incoming Request
    │
    ├─ Load applicable policies (cached, 30s TTL)
    │
    ├─ Evaluate in order:
    │   1. allowed-chains    → DENY/ALLOW
    │   2. approved-addresses → DENY/ALLOW
    │   3. spending-limit    → DENY/ALLOW
    │   4. rate-limit        → DENY/ALLOW
    │   5. time-window       → DENY/ALLOW
    │   6. auto-approve      → AUTO/QUEUE
    │
    │   ALL must pass. First DENY short-circuits.
    │
    ├── ALLOW → Execute action, log result
    ├── DENY  → Return 403, log denial
    └── QUEUE → Hold for manual approval, notify via webhook

​
Policy Results

Every policy evaluation returns detailed results: 
{
  "approved": false,
  "requiresManualApproval": true,
  "results": [
    { "type": "spending-limit", "passed": true, "reason": "Within daily limit" },
    { "type": "approved-addresses", "passed": true, "reason": "Address whitelisted" },
    { "type": "auto-approve-threshold", "passed": false, "reason": "Value 0.1 ETH exceeds auto-approve threshold of 0.05 ETH" }
  ]
}
These results are stored with every transaction record and returned in API responses, making it easy to debug why a transaction was denied or queued.

​
Approval Queue

Transactions that pass all policies except auto-approve enter the approval queue: 
// List pending approvals
const pending = await fetch(
  `https://api.steward.fi/vault/${agentId}/pending`,
  { headers: { Authorization: `Bearer ${agentToken}` } }
);

// Approve (requires tenant-level auth)
await fetch(
  `https://api.steward.fi/vault/${agentId}/approve/${txId}`,
  {
    method: "POST",
    headers: { "X-Steward-Key": "tenant-key" },
  }
);

// Reject
await fetch(
  `https://api.steward.fi/vault/${agentId}/reject/${txId}`,
  {
    method: "POST",
    headers: { "X-Steward-Key": "tenant-key" },
  }
);
Webhooks are dispatched for approval_required, tx_signed, tx_rejected, and tx_failed events. Configure your webhook URL via the Tenants API.