Skip to main content
Kai Affective Intelligence is designed so raw content never persists. Signals are extracted in memory, aggregated at the cohort level, and delivered via REST API.

​
Signal source

Customer content streams — Slack messages, WordPress comments, survey responses, support tickets. Content is sent to Kai via POST /ingest. Raw text is processed in memory and never written to any database.

​
Kai Brain

Kai’s emotional intelligence framework classifies each item: mood, psychological state cluster, topic, and distress score. The same framework that powers Kai’s AI companion, applied to external content without requiring user history.

​
Pseudonymous uids

Model B customers assign a random uid per user before sending content. Kai tracks signals per uid over time to detect trends and escalation. Kai never holds re-identification keys — the customer does.

​
Aggregation

Supabase Edge Functions aggregate per-uid signals into cohort-level JSON on a nightly schedule and on-demand. Enforces anonymisation floor before writing to cache.

​
Transport

REST API with JSON payloads. Webhook support for real-time escalation alerts via HTTPS POST to customer’s nominated endpoint.

​
Architecture overview

Content streams → POST /ingest → Kai Brain (in-memory) → Aggregation → Cache → GET /pulse / GET /alerts
                                      ↓
                              Webhook alerts (HTTPS POST)

Ingestion

Send content batches or real-time streams for analysis.

Output

Retrieve aggregated mood, state, topic, and action signals.