Overview

The Confidential Agent Protocol defines how a Nada Agent:

  1. receives an input or task,

  2. executes inside a Trusted Execution Environment (TEE),

  3. generates verifiable cryptographic evidence of its execution,

  4. and produces an Execution Receipt that can be verified on an SVM chain.

The protocol is designed to guarantee:

  • Confidentiality — inputs, internal state, prompts, model parameters, and intermediate results remain private inside the TEE.

  • Integrity — the on-chain verifier can confirm the exact binary that executed and the authenticity of the result.

  • Non-interactivity — the TEE produces a receipt that can be verified without extra communication.

  • Compatibility — usable directly from Solana programs with no changes to the SVM runtime.

This document fully describes the execution lifecycle, data structures, algorithms, and verification rules.

Last updated