Skip to main content
This page provides a technical summary of how Zyfai’s agentic stack works as a system. Each component is covered in depth on its own page. The goal here is to show how the layers connect and how a transaction flows from strategy decision to onchain execution.

​
Architecture

Zyfai is structured as a four-layer stack, each with a distinct responsibility:
  • Infrastructure Layer: Account Abstraction foundation
  • Control & Access Layer: permission enforcement
  • Intelligence & Strategy Layer: strategy formation, risk analysis, and market monitoring
  • Execution & Verifiability Layer: onchain execution, rebalancing, and cross-chain settlement
Three types of external actors interact with the stack:
  • Users access Zyfai through the zyf.ai front-end or AI assistants such as OpenClaw
  • Curators manage target registries of approved protocols and pools in the Security Proxy Gateway
  • Builders integrate the agentic yield stack into their own products via the SDK or MCP Server
zyfai-map

​
Infrastructure Layer

Every Zyfai user receives their own Safe7579 Smart Account: a self-custodial smart contract wallet built on Safe and made modular through the ERC-7579 standard. This is the container for all user capital. No shared vaults, no pooled funds. The user’s EOA remains the sole owner. All gas fees are covered by the Pimlico Paymaster, so users never need to hold native tokens. Account deployment, rebalancing, auto-compounding, and every other Agent action is gasless for the end user. This layer is the foundation everything else builds on. Without self-custodial, gas-abstracted smart accounts, autonomous Agent execution is not possible.

​
Control & Access Layer

Once the account exists, the next question is: what is the Agent allowed to do? The Security Proxy Gateway answers this. It is an onchain enforcement framework that validates every Agent transaction at three levels: contract address, function selector, and calldata parameters. Transactions are checked against a curator-managed registry of approved targets. Anything outside the registry is rejected, regardless of what the Agent requests. Permissions are granted via Session Keys, which restrict the Agent to approved contracts and actions in a curator-managed registry. In the Smart Wallet flow, account activation, Session Key authorization for the selected strategy and the first deposit are combined in a single transaction using a pre-deployed Safe7579 Smart Account.

​
The flow:

​
Intelligence & Strategy Layer

With permissions defined and enforced, the Intelligence Engine decides when and where to move capital. Smart Rebalancing v.2 is the core decision engine. It continuously evaluates yield opportunities across all supported chains and protocols. Every potential rebalance must pass six safety checks before it is forwarded for execution. The Intelligence Engine is supported by several components:
  • Quant Strategy Team: develops and maintains the scoring models, risk parameters, and allocation logic that the engine operates on.
  • Security Agents: monitor integrated protocols for anomalies, exploits, and liquidity events. If a protocol triggers a risk flag, the Agent can exit positions proactively.
    • Glider Monitor by Hexens: monitors connected contracts for attacks, newly disclosed vulnerabilities and breaches of configured safety rules. Its alerts can trigger automated exits from affected pools.
  • Risk Dashboard (risk.zyf.ai): provides real-time visibility into pool risk scores, protocol exposure, and the Session Key Registry.
  • Earnings Optimizer: a proprietary algorithmic system that calculates and optimizes yield. It supports partial withdrawals, a feature rarely found in comparable products, and powers Capital Splitting feature.
  • MCP Server / OpenClaw Skill: exposes the Intelligence Layer’s data to external AI agents and tools, enabling Agent-to-Agent commerce through standardized interfaces.
This layer produces a decision: “move X capital from pool A to pool B on chain C.” That decision is then handed to the Execution Layer.

​
Execution & Verifiability Layer

The Execution Engine takes validated strategy decisions and settles them onchain. The engine coordinates several execution mechanisms depending on the action:
  • Pimlico Bundler: bundles and relays ERC-4337 user operations for onchain execution.
  • Auto-compounding: when reward tokens reach the $2 threshold, they are automatically claimed, swapped to USDC / EURC / ETH / BTC, and reinvested into the active position.
  • Capital Splitting: spreads funds across approved pools when the split is expected to improve yield. Users with over $10,000 deposited can also enable derisk splitting to diversify their exposure across multiple pools.
  • OmniAccounts: enables cross-chain atomic rebalancing, powered by an intents-based messaging architecture.
  • Intents: the underlying settlement mechanism for cross-chain execution, ensuring transactions are deterministic, atomic, and MEV-protected.

​
How It All Connects

A simplified end-to-end flow:
  1. The user selects a strategy and confirms a single transaction that activates their Safe7579 Smart Account, authorizes the strategy’s Session Keys and deposits their funds. The Infrastructure Layer handles account setup, while the Control & Access Layer scopes the Agent’s permissions to the selected curator registry.
  2. Once the deposit is credited, the Agent can allocate funds within the selected strategy and protocol settings.
  3. The Intelligence & Strategy Layer continuously monitors markets. When it identifies a yield improvement that passes all six rebalancing checks, it generates an allocation decision.
  4. The Execution & Verifiability Layer checks the transaction through the Security Proxy Gateway and settles it onchain. For cross-chain rebalancing, OmniAccounts coordinates settlement through intents.
The user’s capital is now earning optimized yield. The cycle repeats continuously, 24/7, across all supported chains.
For detailed documentation on each component, navigate to the relevant layer section in the sidebar.