We have all been there: you install an AI assistant promised to highlight "important" emails, only to find your notification tray more cluttered than before. The inevitable result is a total blackout, where you disable all push notifications just to regain a sense of peace. The core problem isn't the definition of importance; it is the failure to distinguish between what is urgent and what is merely loud. Klorn, an AI email assistant currently being refined through four weeks of solo development, approaches this by treating your focus as a finite resource, building an "attention firewall" rather than just another productivity bot.
The Five-Tier Escalation Model
Every notification carries a hidden tax: the cognitive cost of switching tasks. To mitigate this, Klorn implements a five-tier escalation structure that dictates how and when an email reaches you. At the lowest level, Silent, the system logs the mail without alerting the user. The Queue tier aggregates individual messages into a single daily digest. Standard Push delivers traditional smartphone pop-ups, while the Call tier—still in development—is designed to escalate urgent matters to a phone call. Finally, Auto-handle allows the AI to draft responses for routine inquiries, leaving only a confirmation for the user.
This classification is determined by a real-time voting mechanism that evaluates four variables: keyword urgency, past communication history, current calendar status, and a Contact Trust Score. If you are in a scheduled meeting or a deep-work block, the AI automatically pushes incoming mail to lower-priority tiers. The Contact Trust Score is the most critical component, as it is derived from your own behavioral data—such as how frequently you reply to a sender or accept their meeting requests. By relying on these historical patterns rather than manual filter rules, Klorn ensures that the AI’s definition of importance aligns with your actual professional habits.
Building the Real-Time Filtering Stack
Klorn is built on a modern web stack, utilizing Next.js 15 for the frontend and Prisma for database management. The system runs on Postgres, with Supabase session pooling ensuring stability during high-traffic periods. Deployment is handled via Render, and the application integrates with Gmail and Google Calendar using push subscriptions. Unlike traditional polling, which periodically checks for new mail, these push subscriptions trigger an immediate signal from Google’s servers to Klorn, minimizing latency and resource waste.
The intelligence layer is powered by Claude and OpenAI models, leveraging tool-use capabilities to perform actions beyond simple text generation. By granting the AI access to your calendar data, the system can dynamically adjust notification tiers based on your real-time availability. This allows the AI to act as a remote control for your attention, deciding whether a message warrants an immediate interruption or can wait until your next break.
Navigating the Risks of AI-Driven Filtering
While the system effectively manages email, it currently operates in a closed beta with several technical hurdles remaining. The most significant is the lack of an LLM cost guard, which poses a financial risk if email volume spikes, as every tool-use call incurs API fees. Additionally, the system is currently limited to email, meaning it cannot yet control notifications from team messaging platforms like Slack or Jandi. Because the AI learns from your behavior, there is also a risk that it may misinterpret high-frequency communication with a temporary contact as high-priority, requiring users to manually adjust scores during the initial onboarding phase.
As the system matures, the focus remains on refining the precision of the attention firewall rather than expanding feature breadth. For now, Klorn serves as a specialized tool for those whose primary workflow is anchored in email, providing a buffer against the constant noise of the modern digital workspace.
