Truvyx Open Specifications

Truvyx publishes two open, freely licensed specifications for the agentic AI ecosystem. Both are maintained by Truvyx, both are free to implement, and both are designed to be adopted by any team building or evaluating multi-agent AI systems — not just Truvyx customers.

What's Here

Agent Execution Trace Protocol (AETP) →

Current version: v1.0 · Status: Stable

AETP defines the structure a multi-agent AI system should use to record its own execution — every agent's decision, every tool call, every memory operation, every escalation to a human — in a format that any compliant evaluation or diagnostic tool can consume. Think of it as the logging standard for AI agent reasoning.

Truvyx Scenario Language (TSL) →

Current version: v1.0 · Status: Stable

TSL defines a portable format for authoring AI evaluation scenarios — the parameters, constraints, agent roles, and ground truth that make a scenario reproducible on any TSL-compliant platform. A .tsl.json file is a citable, shareable, version-controlled artifact, the way an OpenAPI document is for a REST API.

Why These Exist

Every AI framework currently logs execution differently, and every evaluation scenario currently lives locked inside whichever platform authored it. That means evaluation scenarios can't be shared or cited across teams, and root cause analysis is only possible if you're using whichever framework a given evaluation tool happens to support.

AETP and TSL exist to remove that lock-in. Both specifications are free, both are versioned, and Truvyx maintains open-source reference implementations for both in TypeScript and Python.

License

Both specifications and their reference implementations are licensed under Apache 2.0.

Reference Implementations

PackageRegistrySpec
@truvyx/aetpnpmAETP v1.0
truvyx-aetpPyPIAETP v1.0
@truvyx/tslnpmTSL v1.0
truvyx-tslPyPITSL v1.0

Governance

Truvyx maintains both specifications and accepts public proposals for future versions. Breaking changes are only ever introduced in a new major version (e.g. AETP v2.0); the current stable version of each spec will not change its data model without a version bump.