IntelliBar

Intercepts selected text from any macOS application and sends it to OpenAI/Anthropic/Google models for real-time rewriting with specified tone (casual→professional, verbose→concise) or style modifications. Works by capturing the active text field content via system-level text selection APIs, maintaining the original context, and replacing selected text with model output without requiring copy-paste workflows between windows.

Allows users to submit the same prompt to multiple AI models (OpenAI GPT-4o, Anthropic Claude 3.5, Google Gemini, Perplexity, DeepSeek, etc.) and compare responses side-by-side or sequentially. Implements a provider abstraction layer that normalizes API calls across 8+ different model providers with varying authentication, rate limits, and response formats, enabling users to evaluate model strengths without manual API switching.

Tracks context window limits for each supported model (GPT-4o: 128K, Claude 3.5: 200K, Gemini 2.0: 1M, etc.) and automatically manages prompt/response history to fit within model constraints. Implements context window calculation logic that estimates token counts for user prompts and conversation history, truncating or summarizing older messages when approaching the limit to prevent token overflow errors.

Captures the active text field in any macOS application (email, Slack, code editor, document, etc.) and enables AI-powered editing directly within that field without copy-paste workflows. Uses macOS accessibility APIs to detect the active text field, read selected text, and write modified text back to the original field, maintaining formatting and cursor position where possible.

Captures voice input via macOS native speech recognition (not requiring external services like Whisper by default), converts spoken words to text prompts, and routes them to selected AI models. Integrates with system-level audio APIs to enable hands-free interaction without opening a separate voice recording application or leaving the current workflow context.

Converts AI model responses from text to spoken audio using macOS native text-to-speech (TTS) engine, allowing users to consume AI-generated content audibly without reading. Integrates with the response display pipeline to enable one-click audio playback of any model output, supporting multiple voices and languages depending on macOS TTS capabilities.

Stores all conversation history locally on the user's Mac (not on IntelliBar servers), enabling full-text search across past prompts and responses. Implements a local database or file-based storage system that maintains conversation threads, timestamps, and model metadata, allowing users to retrieve previous interactions without cloud sync or external storage dependencies.

Provides a slash command system (e.g., /code, /es, /5x, /brief) that prepends predefined system prompts or instruction templates to user queries before sending to AI models. Enables rapid switching between common use cases without manually retyping instructions, implementing a lightweight prompt templating system that modifies the effective system prompt based on command selection.

Enables users to run open-source models locally via Ollama (a local LLM runtime) and route prompts to those models through IntelliBar's interface. Also supports 'linking to any other model' via custom API endpoints, allowing integration with self-hosted models, private LLM servers, or alternative model providers not in the built-in list. Implements a provider abstraction that treats local and custom models as first-class citizens alongside cloud providers.

Provides a menu bar icon and keyboard shortcuts (⌘Space for main interface, ⌘K for command menu, ⌘O for instant search) that activate IntelliBar without requiring app switching or window management. Implements system-level keyboard event interception to trigger the interface from any application context, keeping the AI assistant always one keystroke away regardless of active window.

Handles authentication and API key storage for 8+ model providers (OpenAI, Anthropic, Google, Perplexity, DeepSeek, xAI, Meta, local Ollama) through a centralized settings interface. Abstracts provider-specific authentication mechanisms (OAuth, API keys, custom headers) behind a unified configuration system, storing credentials securely (likely in macOS Keychain) and routing requests to the correct provider based on model selection.

Provides full-text search capability (⌘O shortcut) to query across all stored conversations and model responses, returning matching prompts, responses, and metadata (timestamps, model used). Implements a search index or query engine over the local conversation database, enabling rapid retrieval of past interactions without manual scrolling through conversation threads.