Intelligence & coaching · Chatalyze Docs

Layers

Layer	Package / route	Requires model keys?
Deterministic alerts	`@chatalyze/metrics` alerts	No
Revenue recommendations	`recommendActions`	No
Thread analysis (Rules provider)	`RulesModelProvider`	No
Thread analysis (HTTP LLM)	`HttpChatModelProvider`	Yes — all of `MODEL_API_BASE`,`MODEL_API_KEY`, `MODEL_NAME`
Batch run	`POST /api/admin/intelligence/run`	Optional

Structured thread analysis

Model (or rules) output shape:

{
  "stage": "rapport|qualify|offer|objection|close|aftercare|unknown",
  "fan_intent": "string",
  "chatter_quality": 1-5,
  "issues": ["…"],
  "next_best_action": "…",
  "risk": ["…"],
  "revenue_opportunity": "high|med|low",
  "summary": "…"
}

Run intelligence

curl -X POST https://api.chatalyze.co/api/admin/intelligence/run \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "content-type: application/json" \
  -d '{"limit":10}'

# or single session
-d '{"sessionId":"sess_…"}'

Without sessionId, selects up to min(limit ?? 10, 20) open sessions
Persists labels to rm_model_labels and raises durable alerts
Never called from the webhook path

Recommendations

Kinds: fan · chatter · shift ·playbook. Examples: prioritize whale awaiting reply; soft-close when warm with no CTA; escalate off-platform risk.

Scorecard

GET /api/chatters/scorecard

Per chatter: sessions, revenue, conversion rate, SLA breach rate, alert counts, coachPriority =crit×3 + warn×2 + info + slaBreachRate×10.

Session analysis

GET /api/sessions/:id/analysis
# metrics + alerts + labels + recommendations