WidgetAI is a floating desktop widget that gives you instant, friction‑free access to ChatGPT, Claude, and Perplexity from a single, compact terminal‑style interface.
No browser tabs, no copy‑pasting, no context switching: just type a command and get the answer inline.

• One‑click AI access send prompts to ChatGPT, Claude, or Perplexity with simple slash commands.
• Floating always‑on‑top stays above all windows; perfect for developers, researchers, and writers.
• Terminal‑inspired UI classic black/green design with a retro feel, minimal and distraction‑free.
• Automatic fallback if one provider fails, it transparently retries the next in your fallback chain (chatgpt → claude → perplexity).
• Live streaming responses see the AI’s answer appear in real‑time, not just when it finishes.
• Rich formatting renders Markdown tables, code blocks, lists, links, bold, and italic properly (requires markdown).
• Persistent history per provider every prompt and response is saved locally (in ~/.widgetai/history.json).
• Smart memory injection the last 8 successful exchanges are automatically prepended to new prompts, so the AI “remembers” the conversation even after a restart.
• Cross‑model memory switch from ChatGPT to Claude and the previous context is carried over.
• History viewer display recent prompts and answers for any provider with /history [provider].
• Clear history wipe all or per‑provider history with /clear [provider].
• Login assistant if a provider requires you to log in, use /login chatgpt to open the login window and stay authenticated.
• Compact / expanded modes the widget stays small while idle and expands automatically when a response appears.
• Docker support run WidgetAI in a container without installing Python dependencies locally.

WidgetAI is available as a Docker image. Pull and run it with:

docker pull pa9da/widgetai:latest
docker run --rm -it \
  -e DISPLAY=$DISPLAY \
  -v /tmp/.X11-unix:/tmp/.X11-unix \
  -v ~/.widgetai:/root/.widgetai \
  pa9da/widgetai:latest

• Python 3.9 or higher
• PySide6, QtWebEngine, and other dependencies (see requirements.txt)

git clone https://github.com/yourusername/WidgetAI.git
cd WidgetAI
pip install -r requirements.txt
python app.py

1. The widget will appear as a small floating window.
2. Use /login chatgpt (or /login claude, /login perplexity) to log in once – the browser will open; log in and close it. Your session will be remembered.
3. Now you can start asking questions.

All commands are entered in the input field. Press Enter to submit.

WidgetAI stores every user prompt and assistant response in a local JSON file (~/.widgetai/history.json). Each entry is tagged with:

• role – "user" or "assistant"
• provider – "chatgpt", "claude", or "perplexity"
• status – "ok", "pending", "streaming", or "failed"
• timestamp and the actual content.

• When you send a prompt, the system first builds a shared memory string from the last 8 successful exchanges (status "ok") across all providers.
• This memory is prepended to your new prompt as context, so the AI can “remember” what was discussed earlier.
• If you switch from ChatGPT to Claude, Claude will still see the previous conversation history, making the experience seamless.

Use the /history command to see a plain‑text list of recent interactions. The output includes the role, provider, and status (if not ok).

Use /clear to wipe everything, or /clear chatgpt to clear only that provider’s history.

WidgetAI uses the markdown Python library to render responses with full Markdown support.
The following are automatically styled in the output panel:

• Code blocks with syntax highlighting (pre‑formatted)
• Inline code
• Tables
• Unordered and ordered lists
• Bold and italic text
• Links (clickable)
• Blockquotes and headings

You can tweak the UI style, polling intervals, fallback chain, and memory limit by editing core/settings.py.
Common settings include:

• FALLBACK_CHAIN – list of providers to try in order.
• SHARED_MEMORY_LIMIT – number of history entries to use as context.
• COMPACT_WIDTH, COMPACT_HEIGHT, MAX_EXPANDED_HEIGHT – widget dimensions.
• POLL_INTERVAL_MS – how often to check for a new response.

WidgetAI/
├── app.py                     # Main entry point
├── ui/
│   └── main_window.py         # UI logic
├── core/
│   ├── command_router.py      # Parses slash commands
│   ├── web_session.py         # Hidden browser automation
│   ├── provider_base.py       # Base class for AI providers
│   ├── chatgpt_provider.py
│   ├── claude_provider.py
│   ├── perplexity_provider.py
│   ├── provider_registry.py
│   ├── history_store.py       # Persistence & memory management
│   ├── failure_detection.py   # Retry logic
│   ├── output_formatter.py    # Markdown → HTML conversion
│   ├── logging_config.py
│   └── settings.py
├── requirements.txt
└── README.md

Contributions are welcome! If you find a bug or have an idea for a new feature, feel free to open an issue or submit a pull request.

MIT — see the LICENSE file for details.

• Built with PySide6 and Qt WebEngine.
• Inspired by the need for a lightweight AI terminal for daily tasks.

Enjoy your new AI terminal!
If you have any questions, open an issue or reach out.
/chatgpt happy coding!