AEP
中文 EN

Architecture

Layered decoupling and Trusted Infrastructure Services matrix design

Five-Layer Architecture System

AI Application Layer Application layer: autonomous agents, multi-agent workflows, and tool platforms Evidence Capture Layer Capture layer: Evidence MCP, SDK interceptors, and runtime environment probes Evidence Model Layer Model layer: structured evidence, causal events, and machine identity models Cryptographic Layer Cryptographic layer: asymmetric signatures, hash digests, Merkle trees, RFC 3161 timestamps Verification Layer Verification layer: neutral verifiers, offline verification tools, and objective audit reports

Evidence Graph Structure

The evidence graph uses a directed acyclic graph (DAG) to accurately capture the agent execution trail: Events link Artifacts and Identities via generatedBy/used semantics, while precededBy locks causal sequencing.

generatedBy precededBy used used Event 1 Artifact Event 2 Identity

Data Flow

Complete closed loop from action capture to independent verification

  1. AI Agent executes mission-critical tasks and triggers action capture

  2. Evidence MCP captures and standardizes Artifacts and Events

  3. Compute cryptographic hash digests of artifacts and events

  4. Generate asymmetric digital signatures using agent private key

  5. Build causal graphs and construct Merkle integrity proofs

  6. Seal and package into standardized .aep evidence container

  7. Any third party verifies independently offline or online

Evidence Package Container Format

The .aep evidence package is a standardized, out-of-the-box container hermetically sealing manifest specifications, identity metadata, artifact digests, causal event chains, Merkle proofs, and digital signatures.

evidence.aep/
├── manifest.json
├── metadata/
│   └── identity/
├── artifacts/
├── events/
├── evidence_graph/
├── merkle/
└── signature/
    ├── signature.json
    └── signature.pub