Platform architecture

This page is the contract-level and API-level companion to Architecture, which explains the same layers in plain language.

YieldSeeker separates the user, agent runtime, wallet, execution controls, and supported protocols. Agent type determines the runtime goal; the wallet and safety boundary are shared.

flowchart LR
  User[User wallet] --> Account[YieldSeeker account]
  Account --> Agent[Agent type runtime]
  Agent --> Wallet[Isolated Agent Wallet]
  Wallet --> Registry[Adapter Registry]
  Registry --> Adapter[Approved adapter]
  Adapter --> Protocol[Supported protocol]
  Monitoring[Offchain monitoring] -. breach wakes .-> Agent

User and account

A user’s connected wallet owns the account relationship and authorizes user-scoped actions. An account can contain multiple agents where product limits allow it.

Agent runtime

The runtime is type-specific. The current Vault Agent uses Autoseek to evaluate approved vault opportunities; future types can pursue different goals without changing the shared wallet boundary.

Agent intelligence may interpret user preferences or explain decisions, while the type-specific strategy determines what those preferences mean.

Monitoring

Monitoring continuously evaluates active rule boundaries against current vault data. A triggered breach can bypass the normal run-frequency limit and make the agent eligible for its next run. Monitoring wakes the runtime; it does not choose an allocation or execute a transaction.

Agent Wallet

Each agent has an isolated smart wallet that holds its assets and submits validated operations. Read AgentWalletKit integration for the contract-level model.

Execution infrastructure

The wallet does not call arbitrary targets directly. The Adapter Registry maps approved targets to adapters, and adapters validate protocol-specific parameters before interacting with a target.

Source boundaries

API routes expose account, agent, and type-specific operations.
Smart contracts enforce wallet ownership and execution constraints.
Offchain runtimes evaluate data and propose actions.
Supported protocols remain independent contracts with their own risks.