Summary: Giza Protocol is worth cataloging not as just another DeFi agent app, but as a layered control plane for non-custodial autonomous finance. The official docs split the stack into three reusable surfaces: a semantic abstraction layer that translates protocol interactions into MCP-style financial operations, an ERC-4337-based authorization layer that lets users delegate bounded authority to agents without giving up custody, and a decentralized execution layer run by staked operators through an EigenLayer AVS. That makes Giza a useful comparison point for Ritual, Olas, dstack, and smart-account middleware because the key question is not only what agents do, but who defines the semantic interface, who holds delegated authority, who executes intents, and how those layers are governed.
What it does:
Exposes a semantic abstraction layer that turns protocol-specific actions into standardized financial operations for agents
Uses ERC-4337 smart-account infrastructure plus specialized validator modules to let users delegate bounded, programmable authority to agents while keeping asset custody
Runs a decentralized execution layer, described in the docs as an EigenLayer AVS, to handle cross-chain messaging, gas optimization, and atomic batching of agent operations
Splits staking into operator staking for execution-network security and community staking for governance participation
Uses the GIZA token for operator collateral, phased governance participation, and liquidity-pair-based fee processing across multi-token workflows
Positions ARMA as the first live manifestation of the broader stack: autonomous agents optimizing DeFi positions while users remain non-custodial
Key claims:
The core reason Giza belongs in the corpus is the explicit three-layer architecture. The docs separate semantic abstraction, authorization, and execution instead of collapsing everything into one generic AI agents for DeFi story.
The semantic-abstraction layer is a real control surface. Giza says it uses the Model Context Protocol to translate protocol-specific interactions into standardized financial concepts, which means the protocol is partly a language/interface layer for machine actors rather than only an execution network.
The authorization design matters because Giza frames agent autonomy as constrained delegation. The docs emphasize ERC-4337 smart accounts, validator modules, and granular permissions, so the practical governance question is which policies and modules mediate agent power rather than whether the agent is merely non-custodial in branding.
The execution layer makes operator admission explicit. Giza says decentralized execution is implemented as an EigenLayer AVS and that operators stake GIZA as collateral, which turns agent execution into a staked service network with clear admission and accountability surfaces.
The token docs add a useful economic wrinkle: GIZA is not only for staking and governance, but also for AMM-pair liquidity used to process multi-token agent fees. That makes fee routing and liquidity design part of the infrastructure story.
The governance docs are unusually explicit that the system is meant to be council-based and delegation-heavy rather than direct-democracy-first. Specialized technical, economic, and emergency councils therefore become part of the protocol’s control surface, not just an implementation detail during bootstrapping.
Giza cleared the bar because it makes non-custodial autonomous finance legible as four separate but linked layers — semantic abstraction, delegated authority, execution-operator security, and expert governance — which is more analytically useful than filing it as a generic DeFi yield agent or generic crypto-AI brand.
Whitepaper: No standalone Giza Protocol whitepaper surfaced in this pass. The strongest primary materials were the official docs and GitHub repositories collected in ../whitepapers/giza-protocol-primary-sources-2026-05-11.md.