Add Harmonic after your model call.

Harmonic sits between AI output and user-visible or execution-bound continuation.

User → LLM / Agent → Harmonic → allow / constrain / escalate / block → response or execution

const aiOutput = await model.generate(input)

const decision = await harmonic.evaluate({
  response: aiOutput,
  consequence_level: "medium"
})

if (decision.recommended_action === "block") {
  return halt()
}

if (decision.recommended_action === "escalate") {
  return requestReview()
}

return continueWith(aiOutput)

{
  "outcome": "degraded",
  "recommended_action": "constrain",
  "stability_boundary": "partial",
  "artifact_hash": "hs_7f4c..."
}