🔧 Tool call visualization — see what your agent is doing in real-time: colored badges, visible parameters, expandable results. The killer feature missing from every other chat UI.
💬 GPT-like interface — sessions in a sidebar, switch between conversations. Familiar if you've used ChatGPT or Claude.
📋 Multi-session navigation — browse all active sessions including cron jobs, sub-agents, and background tasks
⚡ Live streaming — watch the agent think and write token by token
📊 Token usage tracking — progress bars per session so you know how much context is left
🖼️ Inline images — generated or read images render directly in chat with lightbox preview
🎯 Chat-focused — no settings menus or config panels cluttering the screen. Just the conversation.
🌐 i18n — English and French built-in, easy to extend

🚀 Quick Start

Docker (recommended)

docker run -p 3000:80 ghcr.io/marlburrow/pinchchat:latest

Open http://localhost:3000 and enter your OpenClaw gateway URL + token on the login screen.

Or use Docker Compose:

curl -O https://raw.githubusercontent.com/MarlBurroW/pinchchat/main/docker-compose.yml
docker compose up -d

From source

Prerequisites: Node.js 18+, an OpenClaw gateway running and accessible.

git clone https://github.com/MarlBurroW/pinchchat.git
cd pinchchat
npm install
cp .env.example .env
npm run dev

Optionally edit .env to pre-fill the gateway URL:

VITE_GATEWAY_WS_URL=ws://localhost:18789
VITE_LOCALE=en          # or "fr" for French UI

Production build

npm run build
npx vite preview

Or serve the dist/ folder with nginx, Caddy, or any static file server.

⚙️ Configuration

All configuration is optional — credentials are entered at runtime via the login screen.

Variable	Description	Default
`VITE_GATEWAY_WS_URL`	Pre-fill the gateway URL on the login screen	`ws://<hostname>:18789`
`VITE_LOCALE`	UI language (`en` or `fr`)	`en`

Note: The gateway token is entered at runtime and stored in localStorage — it is never baked into the build.

🏗 Architecture

graph TD
    subgraph Browser["🌐 PinchChat (Browser)"]
        Login["LoginScreen<br/><i>credentials</i>"]
        App["App.tsx<br/><i>router</i>"]
        UI["Chat + Sidebar<br/><i>main UI</i>"]
        Hook["useGateway<br/><i>WebSocket state machine</i><br/>auth · sessions · messages"]

        Login --> App --> UI
        App & UI --> Hook
    end

    Hook <-->|"WebSocket (JSON frames)"| Gateway["🔌 OpenClaw Gateway<br/><code>ws://host:18789</code>"]
    Gateway <-->|API| LLM["🤖 LLM Provider<br/><i>Anthropic, OpenAI, etc.</i>"]

Key Components

File	Role
`src/hooks/useGateway.ts`	WebSocket connection, auth, message streaming, session management
`src/components/LoginScreen.tsx`	Runtime credential entry (stored in `localStorage`)
`src/components/Chat.tsx`	Message list with auto-scroll and streaming display
`src/components/ChatInput.tsx`	Input with file upload, paste, drag & drop, image compression
`src/components/ChatMessage.tsx`	Markdown rendering, tool calls, thinking blocks
`src/components/Sidebar.tsx`	Session list with token usage bars and activity indicators
`src/components/Header.tsx`	Connection status, token progress bar, logout
`src/lib/i18n.ts`	Lightweight i18n (English + French)
`src/lib/gateway.ts`	WebSocket protocol helpers and message types

Data Flow

Login — User enters gateway URL + token → stored in localStorage
Connect — useGateway opens a WebSocket and authenticates with the token
Sessions — Gateway pushes session list; user selects one in the sidebar
Messages — Messages stream in via WebSocket frames; the hook assembles partial chunks into complete messages
Send — User input (+ optional file attachments) is sent as a JSON frame over the WebSocket

🛠 Tech Stack

React 19
Vite 7
Tailwind CSS v4
Radix UI primitives
highlight.js via rehype-highlight
Lucide React icons
react-markdown with GFM

📄 License

📋 Changelog

See CHANGELOG.md for a detailed history of changes.

🤝 Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines.

🔗 Links

OpenClaw — the AI agent platform PinchChat connects to