Pointer

A secure AI chat application crafted for deep thinkers

中文 | English

Pointer is a cross-platform desktop AI chat client built with Electron 35, React 19 and TypeScript. It talks to any OpenAI-compatible API (OpenAI, DeepSeek, compatible gateways) and centers on two things: conversation branches kept as a tree, and data you can read yourself — every conversation is a Markdown file inside a workspace folder.

The kernel runs in the main process and is the single source of truth: it owns persistence, the SQLite index, sessions, AI streaming, search and attachments. The renderer holds view state plus a mirror of kernel state fed by events, and never touches files or paths. See docs/architecture.md for the layering, data flow and storage layout.

• Configure multiple LLMs (base URL, API key, model name) and reusable model parameter configs (system prompt, temperature, top-p)
• Pick the LLM per message, so a conversation can mix models
• Streaming responses with reasoning content when the provider reports it

• Tree-structured message history: regenerate or continue creates sibling branches
• Switch branches and inherit context along the selected path
• Outline and topics for long conversations, plus in-conversation search

• Folder hierarchy that maps to real directories on disk
• Starred pages and a multi-tab workspace for parallel conversations
• Workspace-wide search with keyword highlighting, filtered by role, time range and folder
• Message queue per conversation: queued prompts are sent automatically while the conversation is idle

• Import ChatGPT (OpenAI export), DeepSeek and Pointer data
• Export through a source × format matrix (messages, text snippet, table block, code block → Markdown, TXT, HTML, CSV) with preview, editing and screenshot copy
• Multiple accounts and workspaces, each workspace an ordinary folder that can be synced or committed
• All data stays local

CI publishes a Windows installer on the Releases page:

• Windows: pointer-<version>-setup.exe

macOS and Linux packages are configured but not published automatically yet: pnpm build:mac and pnpm build:linux produce them locally, while the release workflow currently builds Windows only.

Requirements:

• Node.js >= 22.12 (required by Vitest 5, which runs the test suite; the app itself runs on the Node bundled with Electron 35)
• pnpm

# Clone the repository
git clone https://github.com/experdot/pointer.git
cd pointer

# Install dependencies
pnpm install

# Run in development mode
pnpm dev

pnpm dev            # development server with hot reload
pnpm start          # run the production build locally
pnpm build          # type check + build
pnpm build:win      # Windows build
pnpm build:mac      # macOS build
pnpm build:linux    # Linux build
pnpm build:unpack   # build without packaging
pnpm lint           # ESLint (warnings fail the run)
pnpm format         # Prettier
pnpm typecheck      # type check main + renderer
pnpm test           # unit tests (domain + renderer mirror)
pnpm smoke:kernel   # boot the kernel in a temp dir and verify the query/command/event chain
pnpm release        # build and publish to GitHub Releases

1. Launch the application and open Settings
2. Add an LLM configuration under "LLM 配置":
   • Name: display name for the configuration
   • API endpoint: an OpenAI-compatible base URL (e.g. https://api.openai.com/v1)
   • API key: your access token
   • Model: model identifier (e.g. gpt-4o, deepseek-chat)
3. Optionally add a model parameter config (system prompt, temperature, top-p) and set the defaults
4. Test the connection from the same panel

• Application data (accounts, session, per-account settings) lives in the Electron user data directory.
• Every workspace is an ordinary folder: pages are <folder>/<page>.pointer.md — Markdown you can open in any editor or commit to git — while kernel-private data lives in .pointer/ (folder metadata, view state, SQLite index, attachments, message queues).
• The SQLite index is derived data: delete it and it is rebuilt by rescanning the workspace.
• Edits made to page files outside the app are detected and reflected in the UI.
• Custom workspace directories are supported, so conversations can live in a folder you already back up or sync.

src/
├── main/                 # Electron main process
│   ├── index.ts          # window, kernel assembly, IPC wiring
│   ├── runtime/          # kernel: persistence, SQLite index, sessions, AI streaming, search
│   ├── ipc/              # runtimeIpc.ts (renderer entry), windowIpc.ts
│   ├── runtimeHost.ts    # dialogs and system integration injected into the kernel
│   └── autoUpdater.ts
├── preload/              # context bridge: runtime / electronWindow / shortcuts
├── renderer/src/         # React UI (components, stores, services, hooks, runtime mirror)
└── shared/
    ├── domain/           # pure entities, invariants and algorithms
    └── contracts/        # RuntimePort: queries, commands, events

• Architecture — layering, data flow, contracts, storage layout
• Auto-update guide（中文）
• Windows code signing guide（中文）

1. Fork this repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Follow the code style (TypeScript, ESLint, Prettier) and run pnpm lint && pnpm typecheck && pnpm test before committing
4. Commit using conventional commits format
5. Open a Pull Request

MIT License