Managing Agents

Overview

Agents are autonomous executors that can perform transactions on behalf of wallets. They operate with capability-gated permissions, budget controls, and support both autonomous and supervised execution modes.

Agent Lifecycle

Agents move through these states:

Creating Agents

Basic Agent

npm run cli -- agent create my-trader \
  --wallet-id <walletId> \
  --mode autonomous \
  --intents transfer_sol query_balance

With Budget

npm run cli -- agent create yield-farmer \
  --wallet-id <walletId> \
  --mode supervised \
  --intents swap stake unstake query_positions

curl -X POST http://localhost:3000/api/v1/agents \
  -H "Content-Type: application/json" \
  -H "x-api-key: dev-api-key" \
  -d '{
    "name": "trading-bot",
    "walletId": "<walletId>",
    "executionMode": "autonomous",
    "allowedIntents": ["transfer_sol", "query_balance"],
    "budgetLamports": 10000000
  }'

const agent = await client.agent.create({
  name: 'trading-bot',
  walletId: walletId,
  executionMode: 'autonomous',
  allowedIntents: ['transfer_sol', 'swap', 'query_balance'],
  budgetLamports: 10_000_000
});

console.log(`Agent ID: ${agent.id}`);
console.log(`Status: ${agent.status}`);

Execution Modes

Autonomous Mode

Agent executes transactions automatically based on its decision engine:

{
  "name": "auto-trader",
  "executionMode": "autonomous",
  "autonomy": {
    "enabled": true,
    "mode": "execute",
    "cadenceSeconds": 30,
    "maxActionsPerHour": 60,
    "steps": [
      {
        "id": "step-1",
        "type": "swap",
        "protocol": "jupiter",
        "intent": {
          "inputMint": "So11111111111111111111111111111111111111112",
          "outputMint": "EPjFWdd5AufqSSqeM2qN1xzybapC8G4wEGGkZwyTDt1v",
          "amount": "1000000",
          "slippageBps": 50
        },
        "cooldownSeconds": 30,
        "maxRuns": 100
      }
    ],
    "rules": [
      {
        "id": "rule-1",
        "when": [
          {
            "metric": "balance_lamports",
            "op": "lt",
            "value": 1000000
          }
        ],
        "then": {
          "type": "query_balance",
          "protocol": "system-program",
          "intent": {}
        },
        "cooldownSeconds": 60
      }
    ]
  }
}

Supervised Mode

Agent requires explicit approval for each transaction:

{
  "name": "supervised-bot",
  "executionMode": "supervised",
  "allowedIntents": ["swap", "stake"]
}

In supervised mode, all transactions pause at approval_gate regardless of policies.

Setting Capabilities

Control which intents and protocols an agent can use:

npm run cli -- agent caps-set <agentId> \
  --intents transfer_sol swap stake query_balance \
  --mode autonomous

Capability Manifests

Manifests are signed, time-limited capability certificates:

Issue Manifest

Create a signed capability manifest:

npm run cli -- agent manifest-issue <agentId> \
  --intents transfer_sol swap \
  --protocols system-program jupiter \
  --ttl 3600

Response:

{
  "manifestId": "manifest-uuid",
  "agentId": "agent-uuid",
  "allowedIntents": ["transfer_sol", "swap"],
  "allowedProtocols": ["system-program", "jupiter"],
  "issuedAt": "2026-03-08T12:00:00.000Z",
  "expiresAt": "2026-03-08T13:00:00.000Z",
  "signature": "manifest-signature"
}

Verify Manifest

Verify a manifest’s authenticity:

npm run cli -- agent manifest-verify <agentId> \
  --manifest '{"manifestId":"...","signature":"..."}'

Or from file:

npm run cli -- agent manifest-verify <agentId> \
  --manifest-file manifest.json

Execute with Manifest

The agent runtime enforces manifest permissions at execution time.

Manifests require AGENT_MANIFEST_SIGNING_SECRET and AGENT_MANIFEST_ISSUER configured. Set AGENT_REQUIRE_MANIFEST=true to enforce manifest validation.

Budget Operations

Manage agent spending budgets:

Check Budget

npm run cli -- agent budget <agentId>

Response:

{
  "agentId": "agent-uuid",
  "budgetLamports": 10000000,
  "spentLamports": 2500000,
  "remainingLamports": 7500000,
  "lastResetAt": "2026-03-08T00:00:00.000Z"
}

Allocate Budget

npm run cli -- treasury allocate \
  --target-agent-id <agentId> \
  --lamports 5000000 \
  --reason "Monthly allocation"

Rebalance Between Agents

npm run cli -- treasury rebalance \
  --source-agent-id <agentA> \
  --target-agent-id <agentB> \
  --lamports 1000000 \
  --reason "Rebalance budgets"

Agent Lifecycle Management

Start Agent

npm run cli -- agent start <agentId>

Stop Agent

npm run cli -- agent stop <agentId>

Pause Agent

npm run cli -- agent pause <agentId> --reason "Manual review"

Resume Agent

npm run cli -- agent resume <agentId>

Executing Agent Transactions

Manually trigger agent execution:

npm run cli -- agent exec <agentId> \
  --type swap \
  --protocol jupiter \
  --intent '{
    "inputMint": "So11111111111111111111111111111111111111112",
    "outputMint": "EPjFWdd5AufqSSqeM2qN1xzybapC8G4wEGGkZwyTDt1v",
    "amount": "1000000",
    "slippageBps": 50
  }'

Autonomy Configuration

Configure autonomous decision-making:

Decision Rules

Rules trigger actions based on conditions:

{
  "rules": [
    {
      "id": "low-balance-alert",
      "when": [
        {
          "metric": "balance_lamports",
          "op": "lt",
          "value": 1000000
        }
      ],
      "then": {
        "type": "query_balance",
        "protocol": "system-program",
        "intent": {}
      },
      "cooldownSeconds": 60
    }
  ]
}

Supported operators: lt, lte, gt, gte, eq, ne Supported metrics:

balance_lamports
token_balance
position_value
daily_pnl
price_deviation

Strategy Steps

Steps define recurring autonomous actions:

{
  "steps": [
    {
      "id": "daily-rebalance",
      "type": "swap",
      "protocol": "jupiter",
      "intent": {
        "inputMint": "So11111111111111111111111111111111111111112",
        "outputMint": "EPjFWdd5AufqSSqeM2qN1xzybapC8G4wEGGkZwyTDt1v",
        "amount": "1000000",
        "slippageBps": 50
      },
      "cooldownSeconds": 86400,
      "maxRuns": 30
    }
  ]
}

Paper Trading & Backtesting

Execute Paper Trade

Test strategies without real funds:

npm run cli -- strategy paper-execute \
  --agent-id <agentId> \
  --wallet-id <walletId> \
  --type swap \
  --protocol jupiter \
  --intent '{...}'

Backtest Strategy

Replay historical strategy:

npm run cli -- strategy backtest \
  --wallet-id <walletId> \
  --name "Q1 Strategy" \
  --steps '[
    {
      "type": "query_balance",
      "protocol": "system-program",
      "intent": {},
      "timestamp": "2026-01-01T00:00:00.000Z"
    },
    {
      "type": "swap",
      "protocol": "jupiter",
      "intent": {...},
      "timestamp": "2026-01-02T00:00:00.000Z"
    }
  ]' \
  --minimum-pass-rate 0.7

List Paper Trades

npm run cli -- strategy paper-list <agentId>

Real Examples from Source

From `scripts/devnet-multi-agent.ts`

import { createAgenticWalletClient } from '@agentic-wallet/sdk';

const client = createAgenticWalletClient('http://localhost:3000', {
  apiKey: 'dev-api-key'
});

// Create wallet
const wallet = await client.wallet.create({ label: 'multi-agent-wallet' });

// Create agent
const agent = await client.agent.create({
  name: 'trader-bot',
  walletId: wallet.id,
  executionMode: 'autonomous',
  allowedIntents: ['transfer_sol', 'query_balance'],
  budgetLamports: 5_000_000
});

// Execute transaction
const tx = await fetch(`http://localhost:3000/api/v1/agents/${agent.id}/execute`, {
  method: 'POST',
  headers: {
    'content-type': 'application/json',
    'x-api-key': 'dev-api-key'
  },
  body: JSON.stringify({
    type: 'transfer_sol',
    protocol: 'system-program',
    intent: {
      destination: destinationPubkey,
      lamports: 1_000_000
    }
  })
});

Agent Security

Security Best Practices:

Always set budgetLamports to limit agent spending
Use supervised mode for high-value operations
Restrict allowedIntents to minimum necessary
Enable AGENT_REQUIRE_MANIFEST=true in production
Set AGENT_REQUIRE_BACKTEST_PASS=true to require strategy validation
Use capability manifests with short TTL for time-limited operations

Governance Controls

# Environment configuration
AGENT_MANIFEST_SIGNING_SECRET=your-secret
AGENT_MANIFEST_ISSUER=your-org
AGENT_REQUIRE_MANIFEST=true
AGENT_REQUIRE_BACKTEST_PASS=true
AGENT_PAUSE_WEBHOOK_SECRET=webhook-secret

Usage Guides

Advanced

Overview

Agent Lifecycle

Creating Agents

Basic Agent

With Budget

Execution Modes

Autonomous Mode

Supervised Mode

Setting Capabilities

Capability Manifests

Budget Operations

Check Budget

Allocate Budget

Rebalance Between Agents

Agent Lifecycle Management

Start Agent

Stop Agent

Pause Agent

Resume Agent

Executing Agent Transactions

Autonomy Configuration

Decision Rules

Strategy Steps

Paper Trading & Backtesting

Execute Paper Trade

Backtest Strategy

List Paper Trades

Real Examples from Source

From `scripts/devnet-multi-agent.ts`

Agent Security

Governance Controls

Next Steps

Protocol Interactions

Setting Policies

Build docs developers (and LLMs) love

Usage Guides

Advanced

​Overview

​Agent Lifecycle

​Creating Agents

​Basic Agent

​With Budget

​Execution Modes

​Autonomous Mode

​Supervised Mode

​Setting Capabilities

​Capability Manifests

​Budget Operations

​Check Budget

​Allocate Budget

​Rebalance Between Agents

​Agent Lifecycle Management

​Start Agent

​Stop Agent

​Pause Agent

​Resume Agent

​Executing Agent Transactions

​Autonomy Configuration

​Decision Rules

​Strategy Steps

​Paper Trading & Backtesting

​Execute Paper Trade

​Backtest Strategy

​List Paper Trades

​Real Examples from Source

​From scripts/devnet-multi-agent.ts

​Agent Security

​Governance Controls

​Next Steps

Protocol Interactions

Setting Policies

Build docs developers (and LLMs) love

Overview

Agent Lifecycle

Creating Agents

Basic Agent

With Budget

Execution Modes

Autonomous Mode

Supervised Mode

Setting Capabilities

Capability Manifests

Budget Operations

Check Budget

Allocate Budget

Rebalance Between Agents

Agent Lifecycle Management

Start Agent

Stop Agent

Pause Agent

Resume Agent

Executing Agent Transactions

Autonomy Configuration

Decision Rules

Strategy Steps

Paper Trading & Backtesting

Execute Paper Trade

Backtest Strategy

List Paper Trades

Real Examples from Source

From `scripts/devnet-multi-agent.ts`

Agent Security

Governance Controls

Next Steps