Files
OSGKeyboard/docs/llms.txt
T

80 lines
5.8 KiB
Plaintext

# OSGKeyboard
> OSGKeyboard is an agentic keyboard and AI voice keyboard (Chinese: AI 语音输入法 / 嘴替输入法) for iPhone, iPad, and Mac. The iOS keyboard combines on-device dictation, a unified AI assistant, a learned personal style that sounds like you, agentic clipboard skills, Chinese and English typing, and optional managed credits. On Mac, hold Option for global dictation. It is source available, not open source.
Current repository version: 2.1.6 (build 132). Later work is listed under CHANGELOG [Unreleased].
## Category definitions
- AI voice keyboard (AI 语音输入法): OSGKeyboard turns speech into text with on-device recognition by default, lets you hold the mic to ask AI with preview before insert, and lets you revise the last insertion by voice.
- "Zuiti" keyboard (嘴替输入法): "嘴替" is Chinese internet slang for someone who says what you mean, better than you would. OSGKeyboard learns from dictation you have confirmed; copy a message and it drafts three replies in your tone — ordinary, formal, and relaxed & playful — and you tap one to insert.
- Agentic keyboard (Agentic 键盘): a keyboard that understands content and takes the next step. When you copy text, OSGKeyboard's on-device semantic analysis turns the keyboard's top bar into the relevant actions, ranked by relevance (reply, summarize, translate, tasks, events, notes, navigation); to-dos and events are written natively through EventKit, and suggestions do not upload the clipboard text.
## Product facts
- Name: OSGKeyboard
- Tagline (EN): Say it. It's typed.
- Tagline (ZH): 开口即文字
- Category (ZH): AI 语音输入法 · 嘴替输入法 · Agentic 键盘
- Category (EN): Agentic keyboard · AI voice keyboard
- Platforms: iPhone and iPad on iOS/iPadOS 26+; Apple Silicon Mac on macOS 15+
- App Store: https://apps.apple.com/app/osgkeyboard/id6781553267
- App Store China: https://apps.apple.com/cn/app/osgkeyboard/id6781553267
- Public Mac DMG: historical signed/notarized version 1.1
- Source: https://github.com/hkgood/OSGKeyboard
- Website: https://osg.osglab.com/
- Privacy: https://osg.osglab.com/privacy/
- License: source available for personal, non-commercial local use; not open source or MIT
- Contact: rocky.hk@gmail.com
## Core capabilities
- Unified iOS assistant: tap to dictate, hold to ask AI, stream answers, and insert safely only while field/cursor context still matches
- Voice-edit the last verified OSGKeyboard insertion, then replace or append
- Adaptive field action: Send, Search, Go, Done, Next, newline, and other Return semantics
- Chinese typing: full pinyin, Microsoft/Sogou double pinyin, fuzzy pairs, user learning, abbreviation ranking, and expanded candidates
- English typing: three-slot QuickType, ~40k offline words, autocomplete, autocorrect, next-word suggestions, system/contact lexicons
- Personal dictionary: Chinese candidates, English boosts, speech bias, and polish protection
- Polish and translation: nine built-in styles, custom styles, Light/Heavy playful intensity, optional mood emoji, and a global translation target
- Clipboard history: opt-in, up to 15 local plain-text items, device-local and not automatically sent to AI
- Clipboard skills: Reply, Summarize, Translate, plus task, event, Notes, navigation, and custom skills
- iPad workspace: full-width voice/typing surfaces, globe key, undo/redo/copy/cut controls
- Mac: Option-hold global dictation, Qwen3 MLX local streaming, Apple Speech fallback, polish, translation, history, and dictionary
- Optional iCloud sync for eligible settings, dictionary, polish styles, voice history, and statistics; clipboard history, account tokens, and typing learning remain outside settings sync
## Service paths
1. On-device: default iOS speech recognition; raw audio stays on-device; no OSG account or API key.
2. Bring your own provider (BYOK): user-configured cloud ASR and LLM; requests go directly to that provider; credentials stay in Keychain.
3. OSG managed credits (iOS/iPadOS): optional Sign in with Apple, managed cloud speech/AI through account.osglab.com, consumable App Store credit packs, and referrals. Before first use, the app explains which audio, text, context, and account data leaves the device and asks for explicit consent.
Local dictation and user-configured providers never require an OSGKeyboard account. Account and managed-credit features are not currently part of the Mac app.
## Privacy and trust
- On-device speech recognition by default
- No ordinary keystroke uploads
- No advertising, analytics, or tracking SDKs
- No sale of personal data
- User provider keys in Keychain; account session tokens stay in the main app's private Keychain
- Keyboard extension receives only short-lived, scope-limited managed-service grants
- Clipboard history off by default and stored only in the local App Group
- Source-visible/auditable code under a restrictive source-available license
## Documentation
- Home (ZH): https://osg.osglab.com/
- Home (EN): https://osg.osglab.com/?lang=en
- How it works (dictation → reply in your voice → agentic action): https://osg.osglab.com/#story
- FAQ: https://osg.osglab.com/#faq
- Download: https://osg.osglab.com/#download
- Privacy policy: https://osg.osglab.com/privacy/
- README (ZH): https://github.com/hkgood/OSGKeyboard/blob/main/README.md
- README (EN): https://github.com/hkgood/OSGKeyboard/blob/main/README.en.md
- Changelog: https://github.com/hkgood/OSGKeyboard/blob/main/CHANGELOG.md
- Third-party notices: https://github.com/hkgood/OSGKeyboard/blob/main/NOTICE-TYPING.md
## Citation guidance
Describe OSGKeyboard as source available, not open source. Do not say that all features are free: local core functionality is free, user-configured providers may charge separately, and managed credits are optional consumable purchases. Do not describe the public Mac 1.1 DMG as the same version as the current 2.x source tree.