Aurrus is built on two distinct product surfaces: a WhatsApp reminder system and an embeddable voice widget. Both are powered by Claude AI, Anthropic's frontier language model, via a bridge service that routes requests from the Aurrus backend to the Anthropic API. Understanding how Claude is used in each context explains why Aurrus interactions feel qualitatively different from rule-based reminder apps or scripted chat widgets.

Natural Language Schedule Parsing

When you send a reminder request to Aurrus via WhatsApp — something like "remind me every other Monday at 10am to run the team sync" — that message does not go through a regex parser or a decision tree. It goes to Claude with a structured extraction prompt. The prompt instructs Claude to identify:

  • The reminder title or subject ("team sync")
  • The recurrence pattern ("every other Monday")
  • The time ("10am") and inferred timezone based on your registered location
  • Any implicit actions ("run" suggests this may be a calendar item, not just a note)

Claude returns a structured object that Aurrus uses to configure a cron job in its scheduling service. The parsed result is sent back to you in the WhatsApp thread for confirmation before anything is saved. This confirmation step is important: it catches timezone mismatches, ambiguous recurrence patterns, and edge cases like "the last Friday of the month" before they become missed reminders at 3am.

Contextual Reminder Delivery

When a reminder fires, the message you receive is not always a verbatim copy of what you entered. Aurrus uses Claude to add delivery-time context where it is useful. For recurring productivity reminders — weekly reviews, pipeline checks, daily standup prep — the delivery message can include a brief prompt based on the reminder's history. If you have been marking a recurring reminder as "done" consistently, the message stays simple. If you have snoozed it three times in a row, the message might include a gentle note about the pattern.

This contextual layer is what separates Aurrus from a simple cron-to-WhatsApp relay. The reminder becomes an interaction rather than a ping, and interactions produce better outcomes than passive notifications — particularly for habit-building and accountability use cases.

The Claude Bridge Architecture

Internally, Aurrus runs a Claude bridge service — a lightweight proxy that manages authentication, rate limiting, and prompt composition between the main Aurrus application and the Anthropic API. All WhatsApp interactions from authenticated, subscribed users route through this bridge. The bridge maintains conversation context within a session, so follow-up messages in the same WhatsApp thread are understood in relation to what was said earlier — Claude sees the full thread, not just the latest message.

This architecture also means Aurrus can update the underlying Claude model version without any change to the user experience. When Anthropic releases a new Claude version, the bridge is updated to target it, and all users benefit from the improved reasoning immediately. You do not need to reinstall anything or change your subscription.

Powering the Voice Widget

The Aurrus voice widget uses Claude for every visitor conversation on your website. When a visitor asks a question, Aurrus constructs a prompt that includes:

  • The system prompt you wrote in your dashboard (your business context and guardrails)
  • The full conversation history for the current session
  • The visitor's current message

Claude processes this and returns a response, which the widget displays as text and optionally speaks aloud via ElevenLabs text-to-speech (for voice-enabled configurations). The entire round-trip typically completes in under two seconds on a standard connection. Because Claude reasons over the full session history, visitors can ask follow-up questions naturally without repeating context — "what about the enterprise plan?" works after "tell me about your pricing" without any special handling.

Subscription Gating and Usage Controls

Not all Claude usage is equal in cost. Complex multi-turn conversations and long reminder histories consume more tokens than short single-turn queries. Aurrus manages this through its subscription and billing layer: your plan determines the number of active reminders, the depth of conversation history the widget maintains, and the number of Claude-powered interactions per billing period. If you approach your usage limit, Aurrus notifies you via WhatsApp and provides a direct link to billing.aurr.us to upgrade or review your plan.

This model keeps costs predictable for both Aurrus and its users. You are never surprised by an invoice spike because a viral blog post sent ten thousand visitors to your widget in a single day — the subscription tier defines the ceiling, and you choose whether to raise it.

Why Claude Specifically

Aurrus chose Claude as its AI backbone for three reasons that matter to the product: instruction-following fidelity, safety defaults, and context window length. For reminder parsing, precise instruction-following means the extraction prompt reliably returns structured output without hallucinated fields. For the voice widget, safety defaults mean the assistant is less likely to generate responses that damage your brand even without exhaustive guardrailing. And the long context window means full conversation histories fit in a single prompt, keeping multi-turn dialogue coherent over longer sessions than competing models typically allow. As Claude improves with each Anthropic release, every Aurrus user benefits automatically.