Elephas

Elephas integrates at the macOS system level to intercept text input across any application (email, documents, messaging, browsers) and provides real-time writing suggestions, completions, and rewrites without requiring copy-paste workflows. The system uses native macOS accessibility APIs to detect text selection and insertion points, then routes text through an LLM backend (likely Claude or GPT) with application-context awareness to generate contextually appropriate suggestions.

Elephas generates multiple alternative versions of user-selected text with explicit control over tone (formal, casual, friendly, professional), style (concise, detailed, creative), and intent (summarize, expand, explain). This likely uses prompt engineering or fine-tuned LLM instructions to produce consistent stylistic variations without requiring the user to manually craft prompts, with results presented in a comparison UI for quick selection.

Elephas analyzes selected text for grammatical errors, style issues, clarity problems, and readability metrics, then provides inline corrections and explanations. This likely uses a combination of rule-based grammar checking (similar to Grammarly's approach) and LLM-based semantic analysis to catch both mechanical errors and contextual writing issues, with corrections presented as suggestions rather than automatic replacements.

Elephas exposes writing operations (rewrite, expand, summarize, correct, generate alternatives) via customizable keyboard shortcuts that work globally across macOS applications. This likely uses a hotkey listener daemon that intercepts key combinations, captures the current text selection, sends it to the LLM backend, and displays results in a floating panel or popover without interrupting the user's typing flow.

Elephas displays writing suggestions, corrections, and variants in a lightweight floating panel that appears near the cursor or selected text, allowing users to preview results and accept/reject changes without leaving their current application. The panel likely uses macOS native UI frameworks (AppKit or SwiftUI) to render results with minimal visual overhead, and supports quick actions (accept, reject, copy, try another variant) via keyboard or mouse.

Elephas detects which macOS application is active (email client, document editor, messaging app, etc.) and adjusts its writing suggestions to match the expected tone, format, and conventions of that application. For example, email suggestions might prioritize professionalism, while messaging app suggestions might favor brevity and informality. This likely uses application bundle identifiers or window title detection to infer context, then passes this context to the LLM as a system prompt modifier.

Elephas can process multiple text selections or entire documents in sequence, applying the same writing action (rewrite, summarize, correct) to each section and collecting results in a single output view. This likely uses a queue-based architecture where each text segment is processed asynchronously, with results aggregated and presented in a scrollable list or exported format, avoiding the need to manually trigger actions on each paragraph or section.

Elephas integrates with macOS clipboard and text editing APIs to seamlessly accept/reject suggestions, copy results, and replace original text without requiring manual copy-paste. When a user accepts a suggestion, Elephas likely uses the Pasteboard API to copy the new text and then simulates keyboard input (Cmd+V) to paste it into the active application, or uses accessibility APIs to directly modify the text field if available.