Summary OpenLess offers an open-source, local-first voice input solution for macOS and Windows, transforming spoken words into AI-polished text at the cursor. It provides a privacy-focused alternative to commercial SaaS tools by allowing users to integrate their own ASR and LLM credentials. The core benefit is faster, cleaner text generation, especially for crafting sophisticated AI prompts, while ensuring all data remains on the user's machine.

How It Works The app captures audio via a global hotkey, transcribes it using streaming ASR (Volcengine/Whisper-compatible), and polishes the text with a chat-completions LLM (Ark/DeepSeek-compatible). Its key feature is the "Structured" AI prompt mode, which converts loose speech into well-formed prompts for LLMs. The pipeline is Speech → Transcript → Cleanup → Insert at Cursor, with clipboard fallback.

Quick Start & Requirements Download .dmg (macOS) or .exe (Windows) from Releases. Requires macOS 12+ or Windows 10+. Grant Microphone and Accessibility (macOS) permissions. Users must provide their own Volcengine ASR and Ark/DeepSeek LLM API credentials via Settings. See USAGE.md for a walkthrough.

Highlighted Details

• Cross-Platform: macOS (12+) and Windows (10+).
• Open-Source & Local-First: Privacy-centric alternative to SaaS, data stays local.
• BYO Credentials: Integrates user-provided ASR/LLM services, avoiding vendor lock-in.
• AI Prompt Optimization: Dedicated "Structured" mode for precise LLM prompt generation.
• Multiple Output Modes: Raw, light polish, structured, formal.
• Customizable Dictionary: Improves ASR accuracy and LLM context.

Maintenance & Community Sponsored by 悠雾云数据 and jiangmuran. Community via QQ Group: 1078960553. Roadmap includes translation, style memory, and history enhancements.

Licensing & Compatibility MIT license permits broad use, including commercial. Local-first architecture enhances compatibility and privacy.

Limitations & Caveats Requires external ASR/LLM API credentials. macOS Accessibility permissions need an app restart. Tool focuses solely on text generation/polishing, not command execution or direct answering.