Built on a stateless, cloud-native stack designed for infinite horizontal scalability and millisecond latency.
Global CDN & React Widget handling sub-30ms interactions.
Node.js API handling auth, rate-limiting, and context retrieval.
pgvector instance storing millions of embeddings with tenant isolation.
Trace a single message through our stack.
User sends a message. The widget encrypts the payload and attaches the session JWT.
System embeds the query and fetches the top-5 most relevant text chunks from the vector index, filtered by tenant_id.
LLM generates natural response while parallel classifiers check for sentiment and lead signals.
Response streamed to user. If lead detected, data is pushed to CRM via Webhook immediately.