​

Agent

Agent is the conversational orchestration service. It receives messages from channels such as WhatsApp, interprets the user’s intent and calls Platform to perform business actions.

​

Responsibilities

Agent owns:

• Channel webhooks.
• Message normalization.
• LLM calls.
• Intent detection.
• Clarification loops.
• Short-lived conversation state.
• Reply formatting.

Agent does not own durable product state.

​

Channel-agnostic design

Agent should not put business logic inside WhatsApp handlers. WhatsApp-specific code should only receive, validate and map messages into channel-neutral models. The core orchestration should work for future channels such as Telegram or web chat.

​

LLM-agnostic design

Agent should be able to use different LLM providers through adapters. The core should depend on a provider interface, not on Gemini, OpenAI or Claude directly.

​

Redis usage

Redis stores short-lived conversation state:

• Pending clarifications.
• Last interaction context.
• Temporary user context cache.

Redis should not be the only place where durable user facts live.

​

Platform integration

Agent must use Platform internal APIs for business actions:

• Resolve channel identifier to user.
• Fetch user context.
• Register meal entries.
• Register workouts.
• Fetch nutrition summaries.

Platform validates and persists the final result.