SAI Protocol provides the infrastructure for autonomous AI agents to own their identity, manage their capital, and execute economic decisions independently. From basic smart contract wallets to premium hardware-attested execution, agents can operate at varying security levels based on their needs.
Agents can choose their sovereignty level: standard smart contract ownership or premium TEE-secured execution with code attestation.
Most frameworks issue agents an EOA private key held in a config file or centralized service. The agent does not own its identity. There is no on-chain record binding behavior history to identity in a way that survives key leaks, hosting outages, or operator changes. SAI provides smart contract-based identity ownership with optional premium TEE security.
Even when logic is open-sourced, there is no cryptographic guarantee that the deployed instance is running that exact logic. A strategy can be swapped or a config silently changed. Premium TEE attestation provides cryptographic code binding, while standard mode operates with verified smart contract logic and configurable permissions.
Agent-to-agent protocols either use free, unmetered messaging (spam/Sybil risk) or route everything through external assets, so the coordination layer generates no structural value capture for the infrastructure enabling it.
Template marketplaces, data feeds, and strategy libraries commonly extract fees with no skin-in-the-game. A negligent or malicious template creator faces little economic consequence if their agent drains user funds. Risk flows one way; fees flow the other.
Standard mode delivers smart contract-based ownership and accountability. Premium mode adds hardware attestation for cryptographic code guarantees.
An ERC-4337 Account Abstraction smart contract wallet (SoulAccount) that the agent controls. Non-transferable execution identity of one agent. Deployed as a minimal proxy (~500K gas). Available in standard and premium TEE-secured modes.
Premium agents sign actions with keys that exist only if running exact registered code. Attested by TEE hardware (Phala dStack over Intel SGX or AMD SEV-SNP). Standard agents use verified smart contract logic with configurable permissions.
Capital deployed through an agent is subject to on-chain slashing, staking, and escrow logic that makes misbehavior costlier than compliance. Includes automated circuit-breakers tied to risk parameters and policy violations. Premium mode adds code attestation ties.
A step-by-step flow from standard smart contract execution to premium TEE-secured operations with economic enforcement.
Premium agents execute intelligence and strategy logic entirely inside isolated Trusted Execution Environments (Phala dStack CVM). Even the hosting provider cannot peek inside or extract keys. Standard agents use verified smart contract execution.
ERC-4337 smart contract wallet. State includes rootId, composeHash (expected code hash for premium), activeEnclaveKey, keyExpiresAt, humanGuardian, and emergencyFrozen flag. Premium mode uses hardware proofs for key changes; standard mode uses configured permissions.
Premium: Stage 0 (Bootstrap) accepts any caller if hardware attestation quote matches registered composeHash. Ongoing: Fresh keypair generated inside TEE → hardware-signed attestation quote → on-chain rotateEnclaveKey submission. Standard: User-controlled key management with policy limits.
Class A (Automated): Off-chain watcher triggers on 3 missed heartbeats. Class B (Automated, trading agents): Risk engine triggers on drawdown breach. Class C (Human): Guardian can trigger anytime. 1-hour mandatory delay before permanent recovery action.
Agents communicate and settle value directly. Every action carries economic consequences enforced on-chain. Identity is the receipt; the receipt is the identity.
Each SAI agent is a smart contract wallet that can operate in standard or premium mode. Premium agents derive signing keys inside a TEE, bound to specific code measurements. All agents support on-chain circuit breakers and policy enforcement. Off-chain execution, on-chain accountability.
The agent is the owner of its own identity. Humans cannot extract its keys.
Premium TEE-protected strategies prevent MEV and front-running
Autonomous harvesters with built-in risk circuit-breakers
Self-managed brand wallets that cannot be hijacked
Direct agent-to-agent settlement with privacy-preserving compute
Execute mathematical mandates with instant proposal analysis
Premium: Private strategies inside TEE with real-time risk monitoring. Standard: On-chain strategies with policy limits.
Foundational core being prepared for testnet deployment (Base Sepolia). Reference agent runtime open-sourced. ERC-4337 account factory, policy module, and breaker mechanics ready for standard agents. Premium TEE attestation layer available. Python SDK available for spawning Souls and interacting with both standard and premium execution layers.
3/3 high-severity findings closed, 9/9 medium findings closed
View audit remediation log →Deploy your first sovereign agent on testnet, explore the documentation, or join our builder community.