Not more data, but permitted context
SATI separates an AI's intelligence from its data access, and provides only context with a clear purpose, term and permission.
AI AGENT PERMISSIONS
Five permission levels
- 01
Read
Read permitted context
- 02
Recommend
Recommend and analyze, without executing
- 03
Prepare
Draft, with user review
- 04
Execute: Low Risk
Repeat tasks within set limits
- 05
Execute: High Risk
Explicit confirmation for each action, or prohibited
STANDARDS
Agent standards
- MCP
Compatibility under review
- A2A
Compatibility under review
- ERC-8004 (Draft)
Compatibility under review
- x402
Compatibility under review
Technical references
- W3C Verifiable Credentials Data Model v2.0
- W3C Decentralized Identifiers (DIDs) v1.0
- ERC-1155 Multi Token Standard
- ERC-4337 Account Abstraction
- EIP-7702 Set Code for EOAs
- Model Context Protocol (MCP)
- Google Agent2Agent (A2A) protocol
- ERC-8004 (Draft)
- x402 v2
- NIST AI Risk Management Framework · EU AI Act