Skip to main content

Semantic Protocol Reference

The Semantic Protocol links natural-language intents to concrete protocol artifacts. It powers discovery, impact analysis, and safety reviews.

Structure​

  • intent – Natural-language description, category, priority.
  • confidence – Numeric bands describing prediction certainty.
  • vectors – Pre-computed embeddings (512-dim) stored as base64 strings.
  • bindings – URNs referencing data sets, APIs, or agents that satisfy the intent.
  • criticality – Impact tier, review cadence, escalation policy.

Example Manifest​

{
  "protocol": "semantic",
  "intent": {
    "id": "intent.detect-pii",
    "description": "Detect PII leaks across outbound webhooks",
    "category": "governance"
  },
  "confidence": {
    "score": 0.92,
    "explanation": "Validated on 5K webhook payloads"
  },
  "vectors": {
    "model": "text-embedding-3-large",
    "values": "3zvAB..."
  },
  "bindings": [
    { "type": "dataset", "urn": "urn:data:dataset:user_events:v1.1.1" },
    { "type": "api", "urn": "urn:api:service:webhooks:v2" }
  ],
  "criticality": {
    "tier": "high",
    "review_window_days": 14
  }
}

Helper APIs​

import { createSemanticProtocol } from '@cpms/semantic';

const intent = createSemanticProtocol(manifest);
const similar = intent.findSimilar(otherIntent);
const score = intent.criticality();

Similarity helpers compute cosine distance using the stored vectors and return ranked matches for search experiences.