Skip to content

INTEGRATION.md

Latest commit

 

History

History
33 lines (22 loc) · 1.51 KB

INTEGRATION.md

File metadata and controls

33 lines (22 loc) · 1.51 KB

Integration Guide

This guide shows how to integrate the Open Safety Signal Protocol (OSSP) into your AI systems.

1. Model Instrumentation

  1. Identify the AI assets (models, guardrails, datasets) that need safety telemetry.
  2. Implement emitters inside your application services. Each emitter should construct a CloudEvent payload that references an OSSP schema in schema/v1.0.0/.
  3. Ensure every event includes the resource object with model_id and environment.

2. Event Delivery

  • Send events as CloudEvents 1.0 structured JSON (Content-Type: application/cloudevents+json; charset=utf-8).
  • Recommended transport patterns:
    • HTTPS POST to the OSSP collector endpoint.
    • OTLP/HTTP /v1/logs via an OpenTelemetry Collector using the mappings in MAPPINGS.md.

3. Validation

  • Add CI validation similar to .github/workflows/validate.yml to keep example events and documentation aligned with the canonical schemas.
  • Optionally perform runtime validation using the SDKs (see sdk-python).

4. Governance & Compliance

  • Track conformance level (Level A or B) for each emitting system.
  • Maintain governance artifacts (OEPs, policies) in the community repository.
  • Plan for foundation transfer per the Neutrality Pledge.

5. Next Steps

  • Extend the taxonomy with additional domain-specific types under schema/v1.0.0/.
  • Publish per-type mapping guides for other transports (Kafka, Pub/Sub) as needed.
  • Engage with the OSSP Working Group via the community repository.