01 — Agent SDK
Agents are first-class protocol entities.
Build browser-native, wallet-integrated, deterministic agents. Isolation by default. Economics by design.
Agent Manifest
{
"agentId": "research.ai.alpha",
"executionMode": "local | hybrid | network | enterprise",
"requiredStake": 10000,
"capabilities": [
"page_read",
"wallet_simulate",
"data_fetch",
"trade_execute"
],
"maxComputePerSession": 500,
"feeModel": { "type": "per_call", "baseFee": 2, "token": "W3AI" },
"riskCategory": "medium",
"version": "1.0.0"
}Architecture Layers
01
Identity
On-chain agent ID + developer signature
02
Capabilities
Declared scopes the runtime can grant
03
Execution
Sandboxed capsule, deterministic boundaries
04
Permission
User-bound authority, revocable per-session
05
Economics
Fee model, stake, burn impact, payouts
06
Reputation
Slashing history + execution success rate
Execution Capsules
No shared memory. Permission-bridged communication only.
Local
Device-only memory. No network egress.
Hybrid
Delegated compute, encrypted bridge.
Network
Solana-linked settlement & registry.
Enterprise
SLA-bound subnet, audit logs.
Context APIs
- getPageContext()
- getWalletState()
- getVerticalData()
- getStakeTier()
Execution APIs
- simulateTransaction()
- requestPermission()
- executeBoundedAction()
- delegateToCompute()
Economic APIs
- estimateFee()
- checkBurnImpact()
- retrieveAgentRevenue()
- verifyDeveloperStake()
Slashing Logic
Security is economic. Violations cost stake.
Triggers
Permission breach · exploit pattern · malicious output
Penalty
Stake slashed · reputation reduced
Effect
Execution disabled · marketplace removal