Mailyr

Generates complete email drafts by accepting natural language instructions or context snippets, using a fine-tuned language model to compose professional or casual messages without requiring the user to write from scratch. The system likely tokenizes user input, passes it through a transformer-based model (possibly GPT-3.5 or similar), and returns formatted email text directly into Gmail's compose window via browser extension injection.

Transforms existing email drafts across multiple tonal registers (professional, casual, emphatic, formal, friendly) by re-prompting the language model with tone-specific instructions and constraints. The system likely maintains the semantic content while adjusting vocabulary, sentence structure, punctuation, and formality markers through a constrained decoding approach or separate fine-tuned tone-specific models.

Analyzes the current email thread (previous messages, sender, subject context) and suggests contextually appropriate responses by extracting thread history via Gmail API, embedding the conversation context, and passing it to the language model for response generation. The system likely uses a sliding window of recent messages (last 3-5 emails) to maintain coherence while respecting token limits.

Implements a usage-based freemium model where free users receive a daily allowance of AI-generated drafts (likely 5-10 per day) before hitting a paywall, tracked via user authentication and backend quota management. The system uses a simple counter mechanism tied to user ID and calendar day, with upgrade prompts triggered when quota is exhausted.

Deploys Mailyr as a Chrome/Firefox extension that injects UI elements (buttons, sidebars, modals) directly into Gmail's compose interface, enabling draft generation and tone adjustment without leaving Gmail or opening a separate application. The extension communicates with Mailyr's backend API via secure HTTPS requests and uses Gmail's DOM structure to identify compose windows and inject responses.

Generates multiple variations of the same email (typically 2-3 versions) with different approaches, tones, or structures, presenting them side-by-side for user selection. The system likely uses temperature/sampling parameters or separate prompts to create stylistic variation while maintaining semantic consistency, allowing users to compare and choose the best fit.