Airsc miễn phí cho tất cả. Nếu nó giúp bạn, hãy ủng hộ $2–$20.

Ủng hộ
MCP serverAn toànTài liệu & nội dungCần cài đặt chút

stashbase

liliu-z · liliu-z/stashbase

Cung cấp một không gian làm việc viết lách ưu tiên cục bộ, nơi các tác nhân AI như Claude và Codex có thể truy cập các nguồn cá nhân, thực hiện tìm kiếm ngữ nghĩa và đề xuất sửa đổi tài liệu.

An open-source, local-first writing workspace where you can write with Claude Code and Codex, drawing on your own sources and past work.

918 sao31Giấy phép: Apache-2.0Cập nhật 13 ngày trước0 lượt mở

Xem trước

▍

Bạn có thể làm gì với nó

  • Yêu cầu tác nhân AI soạn thảo một bài đăng blog sử dụng các bài viết trước đây của bạn làm hướng dẫn phong cách và tài liệu nguồn.
  • Yêu cầu tác nhân tinh chỉnh một tài liệu kỹ thuật bằng cách tìm kiếm tài liệu nội bộ liên quan và đề xuất cải tiến.
  • Xem lại các chỉnh sửa được đề xuất của tác nhân trên tệp Markdown với tính năng so sánh tài liệu chuyên biệt.

Kiểm tra an toàn

An toàn

Không thấy dấu hiệu rủi ro trong các file đã quét.

Quét tự động các file trong repo. Luôn xem lại code trước khi chạy.

Trích từ README

StashBase

Like an IDE, but for writing.

StashBase is an open-source, local-first writing workspace where you can write with Claude and Codex, drawing on your own sources and past work.

  • 🔎 From files to context. Give your agent fast, accurate search across your sources with text extraction and semantic search.
  • 📝 Doc diff ≠ Code diff. Read edits in the flow of your document, without jumping between old and new lines.
  • ✍️ Make it sound like you. Use your past writing and edits to guide new drafts and preserve your voice.

Product Preview

https://github.com/user-attachments/assets/4ba282a3-85d2-4a08-b7af-562a1e71e716

Document Diff

Review suggested edits directly in the document as you read it. Paragraphs and formatting stay in place, deleted words and phrases are struck through in red, and additions are highlighted in green. Revisions appear within the surrounding prose rather than as a line-by-line code patch.

Accept or reject individual changes, or use Accept All and Reject All to take the whole set. Rejecting everything leaves the file exactly as it was. Ask your Agent to revise an open Markdown document and it proposes rather than overwrites, so nothing reaches the file until you accept a change.

Agent file diffs and editor save-conflict comparisons are separate features and still work the way they did.

Get Started

macOS 12+ (Apple Silicon and Intel) and Windows 10+ (x64) are the primary platforms. Linux x86_64 Debian 12+ / Ubuntu 22.04+ is community-supported. On Intel Macs, local search and PDF/image text extraction require macOS 15+; older supported systems can still open projects and edit files. Agent runtime requirements may vary. Intel installers are available starting with v2.9.5.

On macOS, install with Homebrew:

brew install --cask liliu-z/stashbase/stashbase

Or download the latest release:

PlatformDownload
macOS Apple SiliconStashBase-*-mac-arm64.dmg
macOS IntelStashBase-*-mac-x64.dmg
Windows x64StashBase-*-win-x64.exe
Linux x86_64StashBase-*-linux-amd64.deb or .AppImage

See Installation for platform steps, updates, and troubleshooting.

Start Writing

  1. Enter a project. Open or create a local folder.
  2. Choose an Agent. Select Claude or Codex and complete its setup with your provider account. If you can't use either, the built-in Default Agent is available as a fallback. Sign in to StashBase to use its free credits.
  3. Talk it through. Try: “I want to write a blog post about Humanizer. Interview me to understand what I think about it before drafting.”
  4. Draft and revise. Ask your Agent to draft the article, then review and refine it.

Choose between Chat and Documents modes to suit the task at hand. You can write and revise this way, and use the document diff to review fine edits inside the prose.

See Using StashBase for everyday document and search workflows.

Explore the Gallery

Explore how people organize files and write with Agents. Copy a ready-made project or reuse its prompt to try a workflow with your own material.

For example, How to Start a Startup brings together Stanford CS183B lecture transcripts and a founder playbook.

Explore the Gallery →

Your Files and Your Data

Your documents stay in local project folders. Removing a project from StashBase doesn't delete its files. Browsing, editing, previews, and keyword search work without an account. PDF text extraction and OCR run locally, with required components downloaded on first use.

Agent conversations send your prompts and relevant context to model services. Claude and Codex use your provider accounts; the Default Agent connects through StashBase's hosted model gateway.

Search by meaning requires your own OpenAI or OpenRouter key in Settings → Advanced → Search by Meaning. It sends source text and search queries to your chosen embedding provider. Default Agent credits don't cover this service.

Official builds share basic usage statistics by default, excluding document and conversation content. You can turn this off in Settings → General → Privacy. See Usage statistics for details.

Connect External Agents

Agents in the built-in Chat connect to StashBase's tools automatically.

To use an external MCP client, keep StashBase running and copy the connection configuration from Settings → Advanced → External apps (MCP) into that client. It can read, search, and edit supported files in the project you select.

See MCP Configuration for client examples, transports, and access settings.

Build From Source

Install Node.js 24.11.0+, pnpm, and Python 3.13 with venv support. See Local Development for platform dependencies.

git clone https://github.com/liliu-z/stashbase
cd stashbase
pnpm install
pnpm setup:python

env -u ELECTRON_RUN_AS_NODE pnpm dev

The launch command above uses POSIX shell syntax to clear an inherited ELECTRON_RUN_AS_NODE. On Windows, clear that variable in your shell before running pnpm dev.

See Contributing for OCR setup, testing, and release guidance.

Contributing

StashBase is also an experiment in human-directed, AI-first development. Humans own product direction and trust decisions; AI helps connect design, implementation, review, and evidence. The Project Maintenance Model describes that working loop.

Small focused PRs are preferred. Open an issue before larger changes so scope and direction can be discussed first.

Xem mã nguồn trên GitHub