Specifications
Two specs, both Apache 2.0, both on GitHub. Telemetry defines the event model for content attribution. AIMS defines agent identity. They work independently or together.
Telemetry
v0.1 - Preview
The session-event-outcome model for content attribution in AI agent interactions. Defines how retrieval, citation, engagement, and conversion events are structured, transported, and correlated across multiple observers.
Core event types
| Event | When |
|---|---|
content_retrieved | Content fetched from source by an agent |
content_cited | Content referenced in an agent's response |
content_engaged | User interacted with cited content |
turn_started / turn_completed | Conversation turn lifecycle |
checkout_completed | Purchase confirmed - ties outcome back to content |
Key concepts
- Source roles -
origin,edge,index,agent. Multiple observers can report the same retrieval from different vantage points. - Cross-observer correlation -
the
OA-Telemetry-IDHTTP header links events from different reporters into one corroborated retrieval. - Privacy levels -
four tiers from
full(complete query/response text) tominimal(only content URLs and token counts). No PII. - Multi-session attribution -
prior_session_idslinks research, comparison, and purchase sessions into a full journey.
Protocol bindings
| Binding | Purpose | Spec |
|---|---|---|
| ACP | Embed attribution in Agentic Commerce Protocol checkouts | acp/rfc |
| UCP | Embed attribution in Universal Commerce Protocol checkouts | ucp/EXTENSION |
SDKs
AIMS - Agent Identity and Manifest Standard
v0.2 - Draft
Verifiable identity for AI agents that access web content. AIMS is the identity layer in the four-layer stack - it provides the verified subject that telemetry events are attributed to.
What the manifest declares
| Field | What it declares |
|---|---|
| Identity | W3C DID (any method - did:web recommended) |
| Licences | Content licences held, RSL references, redistribution policies |
| Telemetry | OpenAttribution endpoint this agent reports to |
| Foundation / Deployment | Optional: training data provenance, commercial context |
The agent's DID in the manifest matches the agent_id in telemetry events, linking identity to usage data. Publishers resolve the DID
to verify licences before granting access.
SDK
Python:pip install openattribution-aimsMachine-readable schemas
Both specs have JSON Schema definitions for validation and code generation:
- Telemetry schema.json - canonical session and event schema
- ACP content_attribution schema
- UCP extension schema