Use cases

The E-Model is a pre-processing layer: call it before your LLM (or your routing logic) to decide how to handle a message. Because it is deterministic, sub-5ms and token-free, you can run it on every inbound message without adding latency or cost.

1. Escalate angry or churn-risk messages

Route negative valence with high urgency to a human or a retention flow before the model replies. The response_posture tells you how to frame the reply.

JSON

// "Cancel my account. Your bot is useless."
{
  "annotation": { "emotional_valence": "negative", "urgency_level": "high", ... },
  "routing": { "response_posture": "EMPATHIC_SOLVE", "suggested_action": "acknowledge_then_fix" }
}

2. Prioritize a support queue

Sort inbound tickets by urgency_level and flag critical for immediate handling. No ML infra to run; the score is available synchronously.

JavaScript

const { annotation } = await analyze(msg);
if (annotation.urgency_level === "critical") queue.jump(ticket);

3. Safety net for crisis language

The routing layer returns CRISIS_SUPPORT when unambiguous distress markers are present, so you can surface helplines or hand off to a trained responder instead of a generic model reply.

JSON

// "I don't know what to do, I feel completely alone and scared."
{ "routing": { "response_posture": "CRISIS_SUPPORT", "suggested_action": "escalate_or_resource" } }

4. Adapt tone to cognitive load

When cognitive_load is high or overload, instruct your model to slow down, use shorter sentences, and offer a step-by-step breakdown.

5. Localize for East & Southern Africa

cultural_context flags regional markers (for example asante, eish, lekker) so you can adjust greetings and tone for local users rather than defaulting to a single global voice.