Autonomy demands accountability
As AI systems start taking actions — sending emails, publishing code, approving transactions — companies need a verifiable audit trail of why and how those actions happened. TRACE is the open, model-agnostic protocol for that accountability loop: Action → Policy → Evidence.
Build agents that are not only capable — but accountable.
Why TRACE exists
Modern agents can act — file PRs, send invoices, publish content. But most stacks don’t preserve why an action was taken, which rules were in force, or what proof exists that the right checks ran. TRACE addresses that gap with a vendor-neutral protocol for autonomous computation events.
The core loop
Action → Policy → Evidence
• Action — structured intent (type, actor, target, params, timestamp)
• Policy — constraints (enforce or observe); may require approval
• Evidence — attestable outcomes (checks, artifacts, signatures)The protocol doesn’t dictate storage, identity, or cryptography; it defines the shape of the record so implementations can vary while remaining interoperable.
Principles
- Open by default — Apache-2.0 code, public RFCs, community stewardship.
- Model & vendor agnostic — compatible with any LLM/tool/runtime.
- Human-in-the-loop — approvals and overrides as first-class policy.
- Minimal primitives — small surface; pluggable storage and identity.
- Provability — evidence others can verify independently.
Governance
TRACE Labs stewards the protocol, reference server, and SDKs. Substantial changes land through open RFCs, community review, and versioned specs. Implementation freedom is encouraged; interoperability is required.
Non-goals
- Not a full agent framework or LLM runtime.
- Not a centralized trust authority or a walled garden.
- Not tied to any specific storage, identity, or cloud.
Licensing
Code and reference implementations: Apache-2.0
Specification: CC BY 4.0
Call to action
If your systems take action, they should leave a trail. Adopt the primitives. Propose improvements. Contribute integrations. Build agents that are not only capable — but accountable.