Skip to content
jlbjulioPublic

About

Local AI workspace for Windows. Chat with files, research the web, and generate images, voice, documents and slides. No paid API or account.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

MacBot

A local workspace for chat, files and creative projects. Built for Windows PCs (x64), with no paid model API or required account.

Download portable · Privacy

MacBot

Get started

  1. Download the portable ZIP from Releases and extract it.
  2. Open MacBot.exe. Keep the MacBot support folder beside it.
  3. Let first-time preparation finish. Downloads support pause, resume and retry, and reuse compatible cached files.

You need Windows x64, WebView2, internet for preparation and around 22 GB of free space. Models download once and stay outside the release ZIP. Local chat and creation work offline afterward; web research and remote tools need internet. The executable is unsigned.

What you can do

  • Chat with attached documents, images, audio and video.
  • Research questions on the web with sources and supporting excerpts.
  • Generate images, spoken audio, Word/PDF documents, spreadsheets and slides.
  • Dictate messages, transcribe audio and read replies aloud.
  • Connect local or remote MCP tools in Settings and approve their actions.

Pick Light, Dark or Match Windows in Settings → Appearance. Ctrl+, opens Settings and Ctrl+Shift+O starts a new conversation.

Choose a task beside the message box. Describe the content, style and format you want—for example, “Create a purple and cream proposal as Word and PDF.”

Documents, workbooks and slides adapt their fonts, colours, spacing and layouts to your request.

Files support 20 MB each, up to five per message. Audio supports ten minutes and video five minutes. Video uses sampled frames. PaddleOCR-VL 1.6 reads scans, photos of text and embedded document images locally, in over 100 languages; readable text and tables are extracted directly. Reattaching an unchanged file reuses its processed content.

Long documents are read page by page and indexed in batches. Extraction keeps the complete text; unusually large expanded content is rejected explicitly to protect memory.

Images use Baidu's ERNIE-Image Turbo (Apache 2.0, commercial use allowed) in eight steps. PCs with 16 GB of RAM and an 8 GB NVIDIA GPU get the higher-precision model at 1024 pixels; other PCs use a smaller one at 768 pixels. Exports go up to 2048 pixels; larger exports may be resized. For a transparent PNG or WebP, MacBot renders the subject on a plain background and removes it; busy backgrounds stay opaque. Generated speech is English. Review generated content before using it.

Updating from 0.4.x downloads the new image and file-reading models once (about 9.3 GB, or 11 GB on the larger image profile) and overwrites the previous ones in place.

Your workspace

Your history, attachments, models and generated files stay in MacBot/Data. Back up that folder, or copy it into a fresh portable when updating. Settings → Your data & privacy provides deletion controls.

MacBot has no analytics or automatic conversation uploads. Local history is not encrypted. Research sends searches online; connected tools receive approved arguments. Only trust MCP servers you recognize: local servers run with your Windows permissions. Connections support local commands and Streamable HTTP; browser OAuth and legacy SSE are not included.

The default chat model is Qwen3.5 4B, using Unsloth's Q4_K_M GGUF with llama.cpp. Choose Low, Medium or High thinking in Settings. A PC with 16 GB of RAM and a Vulkan-capable GPU with 4 GB or more is recommended; CPU mode is available. MacBot has its own relaxed, creative personality and a synthetic English voice.

Built with

Area Technologies
Desktop Tauri, Rust, React, TypeScript, Vite and Windows WebView2
Backend Python, FastAPI, Uvicorn, LangGraph and SQLite
Local AI Hugging Face, Unsloth GGUF, llama.cpp, Qwen3.5, EmbeddingGemma 2, Qdrant, PyTorch, Transformers and Sentence Transformers
Image, scanning & voice ERNIE-Image Turbo, stable-diffusion.cpp, PaddleOCR-VL 1.6, Piper and faster-whisper
Files python-docx, ReportLab, openpyxl and python-pptx
Tools & preparation MCP, Node.js and uv

Build the portable

Use Windows x64 with Node.js 22+, Python 3.11, uv, Rust/MSVC, and Visual Studio 2022 Build Tools with C++ and the Windows SDK.

npm ci
$env:UV_PROJECT_ENVIRONMENT = Join-Path $env:LOCALAPPDATA "MacBot/Development/Python"
npm run build:desktop
& "$env:UV_PROJECT_ENVIRONMENT/Scripts/python.exe" scripts/package-portable.py

The desktop build creates or synchronizes the isolated Python environment using backend/uv.lock before preparing the portable. Its log is logs/python-environment.log; existing workspace data is kept outside that environment.

Portable packaging validates uv against the hash in bootstrap/engines.json. It reuses the matching binary in the portable, build/tools/uv.exe, or PATH, so updating global uv does not require downgrading it. A fresh checkout needs the pinned executable in build/tools/uv.exe or a matching uv on PATH.

The ZIP is created in release/. Upload it as the single download in a GitHub Release. Models, personal data and build output are excluded from the source repository.

About

Local AI workspace for Windows. Chat with files, research the web, and generate images, voice, documents and slides. No paid API or account.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages