Inbox Zero

Inbox Zero implements a webhook-based email ingestion system that connects to Gmail and Outlook via OAuth, processing incoming emails in real-time through a webhook handler that parses email metadata, attachments, and content. The system uses provider-specific webhook protocols (Gmail Push Notifications, Outlook Change Notifications) and normalizes them into a unified internal email schema stored in PostgreSQL, enabling immediate processing without polling delays.

Inbox Zero implements a rule engine that allows users to define email automation rules in plain English, which are then parsed by an LLM into structured rule definitions stored in the database. The engine evaluates incoming emails against these rules using semantic matching (not just regex), executing actions like auto-filing, labeling, or blocking based on rule conditions. The system supports rule versioning and A/B testing of rule effectiveness.

Inbox Zero provides a dashboard that tracks email productivity metrics including inbox size over time, reply response times, email volume by category, and rule effectiveness. The system aggregates email metadata and action logs to compute these metrics, and surfaces trends and insights to help users understand their email patterns. Metrics are computed asynchronously and cached to avoid performance impact.

Inbox Zero uses PostgreSQL with a normalized schema that stores emails, conversations, rules, actions, and user profiles. The schema includes tables for email threads (linked via In-Reply-To headers), rule definitions and execution logs, user style profiles, OAuth tokens, and action audit trails. The design supports efficient querying of emails by category, sender, date range, and conversation thread, with indexes optimized for common access patterns.

Inbox Zero analyzes a user's historical email patterns (tone, vocabulary, signature style, response length) and uses this profile to generate contextually appropriate reply drafts for incoming emails. The system extracts user writing style from past sent emails, stores this as a style vector or prompt template, and feeds it to the LLM alongside the incoming email to generate on-brand replies. Users can accept, edit, or regenerate drafts before sending.

Inbox Zero implements a 'Reply Zero' system that tracks which emails require responses and monitors whether replies have been sent. The system uses email threading (In-Reply-To headers, message IDs) to link related emails into conversation chains, marks emails as 'awaiting reply', and surfaces unresponded emails in a dedicated view. It can also auto-generate follow-up reminders for emails that haven't received responses within a user-defined timeframe.

Inbox Zero uses LLM-based content analysis to automatically categorize incoming emails into user-defined categories (e.g., 'urgent', 'promotional', 'meeting request') based on semantic understanding of email content, sender context, and user preferences. The system can extract key information (action items, deadlines, sender intent) and surface this metadata in the UI for quick scanning. Categories can be customized per user and refined over time based on user feedback.

Inbox Zero provides bulk action capabilities (archive, delete, unsubscribe, label) that can be applied to multiple emails at once, with safety features including preview of affected emails, confirmation dialogs, and undo functionality. The system logs all bulk actions with timestamps and user context, allowing users to revert actions within a configurable time window (default 30 days). Actions are executed asynchronously to prevent UI blocking.

Inbox Zero analyzes incoming emails to detect unsolicited cold outreach (sales pitches, recruitment spam) using heuristics like sender reputation (first contact, no prior conversation history), email patterns (generic greetings, promotional language), and sender domain analysis. Detected cold emails can be automatically archived, labeled, or blocked. The system maintains a reputation database of known cold email senders and patterns.

Inbox Zero exposes its email management capabilities through a Model Context Protocol (MCP) server, allowing external AI agents and LLM applications to read emails, execute actions (archive, label, reply), and query email metadata programmatically. The MCP server implements standard MCP tools for email operations and uses OAuth-based authentication to ensure secure access to user email accounts. This enables integration with AI agents that need email context for decision-making.

Inbox Zero analyzes a user's historical email patterns (sent emails, reply patterns, signature style, vocabulary preferences) to build a personalization profile that informs AI-assisted features like reply drafting and email categorization. The system extracts style vectors or prompt templates from historical data and stores them in the database, allowing AI features to be personalized without requiring explicit user configuration. The profile is continuously updated as new emails are sent.

Inbox Zero implements OAuth 2.0 flows for both Gmail and Outlook/Microsoft Graph, handling token acquisition, refresh, and revocation. The system supports multiple email accounts per user, storing OAuth tokens securely in the database with encryption at rest. It manages provider-specific scopes and permissions, and handles token expiration with automatic refresh. The architecture allows users to connect and disconnect email accounts without affecting other accounts.