Gurubot

Delivers real-time conversational AI responses directly within WhatsApp's messaging interface by integrating with WhatsApp Business API, maintaining conversation context across message threads without requiring users to switch applications or manage separate chat windows. The system parses incoming WhatsApp messages, routes them through an LLM inference pipeline, and returns responses formatted for WhatsApp's native text rendering, preserving conversation history within the existing thread structure.

Implements encryption for chat messages using WhatsApp's Signal Protocol (E2EE) combined with server-side encryption for conversation metadata and user profiles, ensuring that message content cannot be intercepted or accessed by Gurubot's infrastructure during transmission or storage. The system leverages WhatsApp's native E2EE for message transport and adds application-layer encryption for any data persisted in Gurubot's backend databases, using AES-256 or equivalent symmetric encryption with key derivation from user credentials.

Delivers AI responses within WhatsApp's messaging interface with minimal perceived latency by implementing response streaming, local inference caching, and connection pooling to WhatsApp's message delivery API. The system likely uses a pre-warmed inference endpoint or edge-deployed model to reduce round-trip time between message receipt and response generation, with streaming tokens sent incrementally to WhatsApp rather than waiting for full response completion before transmission.

Maintains conversation history and user context across multiple message exchanges by storing conversation threads in a backend database indexed by WhatsApp user ID, enabling the AI to reference previous messages and maintain coherent multi-turn dialogue without requiring users to repeat context. The system likely implements a sliding-window context manager that retrieves relevant prior messages from storage, embeds them with the current query, and passes the combined context to the LLM inference pipeline.

Enforces paid subscription tiers by implementing per-user rate limits, message quotas, and feature gating at the API gateway level, where incoming WhatsApp messages are validated against the user's subscription status before routing to the inference pipeline. The system likely maintains a subscription database indexed by WhatsApp phone number, checks quota consumption (messages per day/month), and returns error messages or upgrade prompts when limits are exceeded, preventing free-tier abuse and monetizing the service.

Establishes user identity and account persistence by using WhatsApp phone numbers as unique identifiers, eliminating the need for separate login credentials or account creation flows. The system maps WhatsApp phone numbers to user profiles stored in a backend database, enabling subscription tracking, conversation history retrieval, and personalization without requiring users to create usernames or passwords, leveraging WhatsApp's built-in phone verification.

Tailors AI responses to individual users by retrieving their stored profile data (preferences, conversation history, interaction patterns) and injecting this context into the LLM prompt before generation, enabling the AI to provide personalized advice, remember user preferences, and adapt tone or content style based on prior interactions. The system likely implements a user profile store with fields for preferences, interests, and interaction metadata, which is queried and combined with the current message to create a personalized system prompt or context injection.