Skip to content

Releases: provnai/vex-sdk

VEX SDK v1.6.0 — Protocol Alignment

23 Mar 00:49

Choose a tag to compare

This release aligns the SDK with VEX Protocol v1.6.0 Lockdown and Intent v3 specification. It delivers deterministic cryptographic behavior and structured privacy for AI agent workflows.

🏛️ Core Features

  • HPKE Intent Privacy (RFC 9180)

    • Standardized asymmetric encryption (X25519-HKDF-AESGCM) for tool parameters.
    • Protects PII and API keys within the evidence chain.
    • Enables zero-knowledge intent dispatch to the Gate/Vanguard only.
  • Intent v3 & Authority v3 Realignment

    • Updated structure to include aid (Agent Identity), commands, and intent_context.
    • Enhanced authority signaling with binding_status and continuation_token.
    • Provides a foundation for governance and Action Enforcement Module (AEM) handshakes.
  • 4-Leaf Merkle Tree (RFC 6962)

    • Migrated Evidence Capsule pillars to a Binary Merkle Tree.
    • Domain separation (0x00 leaves / 0x01 internal nodes) for collision resistance.
    • Ensures deterministic root parity across Python and TypeScript SDKs.

🛠️ Developer Experience

  • Hardening (TypeScript)

    • 100% lint clean: 0 errors, 0 warnings.
    • Strict protocol interfaces (VexCapsule, VexResult) to avoid unsafe casts.
    • Patched dependency vulnerabilities using npm audit fix.
  • Local Token Verification

    • Supports offline Ed25519 verification of Gate-issued continuation tokens.
    • Confirms authenticity before tool execution.

📊 Release Summary

  • Version: v1.6.0
  • Protocol Support: Intent v3 / Authority v3 / Binary Merkle Tree
  • Security: HPKE (RFC 9180) / RFC 6962 / 100% Clean Audit

⚡ Quickstart

# Python
pip install provn-vex-sdk --upgrade

# TypeScript
npm install @provnai/vex-sdk@latest

🛡️ VEX SDK v1.5.0

17 Mar 03:26

Choose a tag to compare

The SDK now aligns VEX Protocol and with the Capsule v0.3 specification and provides consistent behavior across both Python and TypeScript implementations.


🏛️ Core Features

Silicon Identity (Attest)

  • Support for hardware-rooted identity bindings
  • Evidence Capsules can be tied to platform identity sources such as:
    • TPM 2.0
    • secure enclave environments
  • This allows capsule signatures to be associated with a verifiable host platform

Capsule Ledger Linking

  • Capsules can now reference the previous capsule through a prev_hash
  • This creates a chained history of agent actions
  • Helps provide a continuous forensic timeline for capsule verification

Authorization Enforcement (ESCALATE)

  • Added protocol handling for the ESCALATE state
  • The SDK now supports authorization flows that require additional approval
  • Includes retry handling and stateful polling for multi-step authorization responses

Cross-Language Parity

  • Python (provn-vex-sdk) and TypeScript (@provnai/vex-sdk) implementations now produce identical capsule outputs
  • Capsules generated in one implementation can be verified in the other with deterministic results

🛠️ Developer Experience

Python Decorators

  • Added @vex_secured decorator
  • Allows simple wrapping of functions to automatically produce Evidence Capsules

TypeScript Middleware

  • Added vexMiddleware for integration with the Vercel AI SDK
  • Allows automatic capsule creation during request handling

Binary Wire Format

  • Added support for the Binary Wire Spec v0x03
  • Enables lower-latency serialization and transport of capsules

Schema Validation

  • Improved schema validation and explicit type exports
  • Additional fixes applied during the internal audit pass

📊 Release Summary

Version: v1.5.0
Protocol Support: Capsule v0.3 / Binary Wire Spec v0x03
Implementations: Python & TypeScript


⚡ Quickstart

# Python
pip install provn-vex-sdk

# TypeScript
npm install @provnai/vex-sdk