Add Harmonic after your model call.
Harmonic sits between AI output and user-visible or execution-bound continuation.
HCS Specification Library
The Harmonic Constitutional Specification (HCS) series defines the implementation-independent constitutional architecture behind continuation governance.
HCS-000 → HCS-001 → HCS-002 → HCS-003 → HCS-004 → HCS-005 → HCS-006 → HCS-099
Flow
User → LLM / Agent → Harmonic → allow / constrain / escalate / block → response or execution
cURL
curl -X POST "https://www.solace-harmonic.com/api/evaluate" \
-H "Authorization: Bearer hs_live_..." \
-H "Content-Type: application/json" \
-d '{"response":"The assistant recommends stopping medication immediately.","consequence_level":"high"}'
Node
async function evaluateWithHarmonic(response) {
const result = await fetch("https://www.solace-harmonic.com/api/evaluate", {
method: "POST",
headers: {
"Authorization": `Bearer ${process.env.HARMONIC_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
response,
consequence_level: "medium",
}),
});
if (!result.ok) throw new Error(await result.text());
return result.json();
}
Python
import os
import requests
def evaluate_with_harmonic(response):
r = requests.post(
"https://www.solace-harmonic.com/api/evaluate",
headers={
"Authorization": f"Bearer {os.environ['HARMONIC_API_KEY']}",
"Content-Type": "application/json",
},
json={
"response": response,
"consequence_level": "medium",
},
timeout=20,
)
r.raise_for_status()
return r.json()
Decision Handling
if (decision.recommended_action === "deny") halt(); if (decision.recommended_action === "escalate") requestHumanReview(); if (decision.recommended_action === "constrain") applySafeCompletion(); if (decision.recommended_action === "allow") continueNormally();
Governance Pack Upgrade API
The Governance Pack bundles four deeper primitives behind one upgrade endpoint: Reality Contact, Authority Continuity, Consequence Boundary, and Runtime Admissibility.
POST /api/governance-pack
Authorization: Bearer hs_live_...
Content-Type: application/json
{
"packet_id": "case-001",
"requested_action": { "type": "clinical_order", "scope": ["care_plan"] },
"declared_reality": {
"current_state_claims": ["Patient status is stable"],
"last_verified_at": "2026-05-26T18:00:00.000Z"
},
"observed_reality": { "signals": [{ "statement": "Recent signals remain stable" }] },
"authority_chain": {
"subject": "clinical-agent-1",
"issuer": "hospital-ops",
"scope": ["care_plan"],
"last_verified_at": "2026-05-26T18:00:00.000Z",
"chain": [
{ "actor": "clinical-agent-1", "status": "active" },
{ "actor": "hospital-ops", "status": "active" }
]
},
"revocation_state": { "last_revocation_check_at": "2026-05-26T18:00:00.000Z" },
"consequence_profile": {
"level": "high",
"reversibility": "reversible",
"execution_surface": "clinical_order"
},
"safeguards": { "operator_review_confirmed": true }
}
Response includes a single package outcome plus primitive-level results and artifact hashes for audit/replay.