Writekin 是一款 macOS 应用,它利用 QLoRA 技术在本地对语言模型进行微调,训练数据来自你自己的邮件、消息和文档,从而让模型能像你一样写作,所有处理都在设备上完成。

Stars

16

7 天增长

暂无数据

Fork 数

0

开放 Issue

0

开源协议

NOASSERTION

最近更新

2026-07-28

AI 仓库情报摘要
FR-AI / ANALYSIS

为什么值得关注

它兼具隐私优先(数据不离机)和个性化写作助手的特性,通过你的真实消息和文件进行训练,并在 Apple Silicon 上借助 MLX 高效运行 QLoRA,是一款值得关注的开源工具。

适合谁使用

  • 重视隐私且希望使用 AI 写作助手但不想依赖云端的写作者
  • 希望写作工具能模仿个人语气的 Apple Silicon Mac 用户
  • 需要生成一致品牌或个人语调内容的创意工作者
  • 对本地微调小语言模型工作流感兴趣的开发者

典型使用场景

  • 从零开始以你的自然语气起草电子邮件或消息
  • 将现有文本改写为匹配你个人风格的内容
  • 创建听起来像你而非通用 AI 的博客文章或日记
  • 为写作任务尝试个性化的本地大语言模型

项目优势

  • 完全隐私:数据摄取、训练和生成均在本地进行,无遥测、无账户
  • 使用用户自己的数据(邮件、消息、文档)微调模型,使其像用户一样写作
  • 借助 MLX 和 QLoRA 在 Apple Silicon(M 系列)上高效训练
  • 以 PolyForm 非商业许可开源,允许验证和个性化定制

使用前须知

  • 仅支持 Apple Silicon Mac,至少需要 16GB 内存(推荐 32GB+ 以训练更大模型),且系统为 macOS 14 或更高
  • 非商业使用免费,商业使用需单独获得许可证
  • 效果取决于用户是否有足够且多样化的个人写作语料

README 快速开始

Writekin

Writing that's kin to yours. Writekin ingests your own writing — Mail, Messages, documents, chat exports — curates it into a training corpus, fine-tunes a local language model (QLoRA on Apple Silicon via MLX), and then drafts and rewrites text that sounds like you. Everything happens on your Mac.

Screenshots

Overview — your corpus at a glance.

Compose — draft and rewrite in your voice.

Train — a local QLoRA run with a live loss chart.

Models — the on-device model library.

Privacy model

  • Nothing leaves your machine. Ingestion, training, and generation are all local. There is no telemetry, no analytics, and no account.
  • The auto-update check (Sparkle) requests one file — the release feed — and sends no system profile. That is the app's only network traffic besides model/tool downloads you explicitly start.
  • Model weights download from Hugging Face on demand; the optional Messages-support tool downloads from its own official GitHub release, checksum-verified. Both happen only when you click.
  • This repository exists so you don't have to take any of the above on faith: the source is the proof.

Requirements

  • Apple Silicon Mac (M-series). 16 GB unified memory minimum; 32 GB+ recommended for training larger models.
  • macOS 14 or newer.
  • Full Disk Access (user-granted) if you want Mail and Messages ingested.

Install

Download the latest DMG from Releases, drag to Applications, open. The app guides you through source detection, Full Disk Access, and model download on first launch.

Messages support uses the open-source imessage-exporter (GPL-3.0), which Writekin downloads on demand — or use an existing brew install imessage-exporter.

Build from source

brew install xcodegen
xcodegen generate
open Writekin.xcodeproj   # build the Writekin scheme

The full test suite (xcodebuild test -scheme Writekin) must pass before submitting changes — see CONTRIBUTING.md (note the contribution licensing terms there).

New to the codebase? Start with ARCHITECTURE.md — the pipeline, the folder map, and the invariants that keep it honest.

License

Writekin is source-available, free for personal (noncommercial) use under the [PolyFo

项目描述

Writing that's kin to yours

相关仓库与替代方案

根据分类、Topic 和编程语言匹配的相似项目。

makecindy
精选
makecindy GitHub avatar

cindy

Cindy is an open-source AI agent that runs locally on your machine, integrates multiple AI harnesses and models, and provides memory, skills, and automation to perform real work in your projects and apps.

AI 与机器学习大语言模型
958
uzairansaruzi
精选
uzairansaruzi GitHub avatar

hermex

Hermex is a native SwiftUI iPhone app that lets you control a self-hosted Hermes AI agent directly from your phone, with no subscriptions, tracking, or third-party relay.

AI 与机器学习大语言模型
941
Blaizzy
精选
Blaizzy GitHub avatar

nativ

Nativ is a native macOS app that lets you run AI models locally on Apple silicon, offering chat, model management, performance analytics, and an OpenAI/Anthropic-compatible API server.

AI 与机器学习AI 视觉与语音
985