OpenAI-compatible GNOME Chat Client

A GNOME-style desktop client for OpenAI-compatible APIs (e.g., OpenAI, local Ollama with OpenAI compatibility). It follows GNOME Human Interface Guidelines (HIG), supports system dark/light mode, and uses your GNOME accent color for primary/UI highlights.

This app provides a clean chat interface with message “bubbles”, a model picker, a settings page for API configuration, and persistence of your settings.

Features

• GNOME-native UI

  • Single HeaderBar with a centered tab switcher (Chat / Settings)
  • Follows system dark/light appearance using GNOME settings
  • Uses your GNOME accent color (suggested-action) for the “Send” button and active tab
  • Minimal CSS only for spacing and rounded message bubbles; theme controls colors

• OpenAI-compatible API

  • Fetch models from /v1/models
  • Send chat completions to /v1/chat/completions
  • Works with OpenAI or any OpenAI-compatible backend (e.g., local Ollama server running OpenAI endpoints)

• Settings & Persistence

  • API Key, Base URL, Default Model
  • Saved under ~/.config/openai_gtk_client/settings.json
  • Settings page separated from chat; saves and returns you to chat

• UX Details

  • Non-blocking network calls via background threads
  • Status InfoBar at the bottom (Ready, Fetching models..., Done, Failed)
  • Chat view with selectable, wrapped text bubbles
  • Keyboard shortcuts

Requirements

• Linux with GNOME (GTK 3 environment)
• Python 3.x
• PyGObject (GTK 3 bindings)
• Requests (HTTP client)

Install dependencies (Debian/Ubuntu-based):

sudo apt update
sudo apt install -y python3-gi gir1.2-gtk-3.0
pip3 install --user requests

On Fedora:

sudo dnf install -y python3-gobject gtk3
pip3 install --user requests

On Arch:

sudo pacman -S --needed python-gobject gtk3
pip3 install --user requests

Installation

Clone or place the project directory where you want. The primary file is:

• ollama_gui/code.py

No build step is required.

Run

From the project root:

python3 ollama_gui/code.py

Optional (for testing dark mode forcibly):

GTK_PREFER_DARK=1 python3 ollama_gui/code.py

Configuration

Settings are accessible via the “Settings” tab in the HeaderBar. They are persisted to:

• ~/.config/openai_gtk_client/settings.json

Fields:

• API Key: Your OpenAI (or other provider) API key. Leave empty for servers that don’t require it (e.g., some local setups).
• Base URL: The base HTTP endpoint (e.g., https://api.openai.com/ or http://localhost:11434/).
• Default Model: Preferred model ID (fetched via the “Fetch Models” button).

Notes:

• The Base URL should end with a slash; the app normalizes this automatically.
• The model list is fetched from /v1/models and parsed to extract id or name fields.

Usage

1. Settings

   • Go to “Settings” in the HeaderBar.
   • Enter your API Key (if required), Base URL, and Default Model.
   • Click “Fetch Models” to populate available models.
   • Click “Save Settings” to persist and return to the Chat tab.

2. Chat

   • In the Chat tab, use the model picker and refresh button to choose a model.
   • Type your message in the multiline input field.
   • Press “Send” to get a response in the conversation area.

3. Status

   • Status messages are shown on the InfoBar at the bottom.

Keyboard Shortcuts

• Ctrl+Enter: Send message
• Ctrl+Comma: Switch to Settings tab

Dark/Light Mode and Accent Color

• The app follows GNOME’s appearance preferences automatically.
• Dark mode:
  • Reads org.gnome.desktop.interface color-scheme (when available).
  • Accepts GTK_PREFER_DARK=1 to force dark (useful for testing).
• Accent color:
  • The “Send” button and the active tab get the system accent via the suggested-action style class.
  • Requires an Adwaita-based theme that supports accent colors (GNOME 45+).

OpenAI-compatible API Expectations

• Models endpoint: GET {base_url}/v1/models
  • Should return either { data: [ { id: "...", ... }, ... ] } or a list with id/name fields.
• Chat endpoint: POST {base_url}/v1/chat/completions
  • Request JSON example:

    {
      "model": "gpt-3.5-turbo",
      "messages": [{ "role": "user", "content": "Hello" }],
      "temperature": 0.7
    }

  • Response is expected to include choices[0].message.content (OpenAI chat format), or fallback to choices[0].text.

Files and Structure

• ollama_gui/code.py — Main application code
• ~/.config/ollama_gui/settings.json — User settings (created on first save)

Troubleshooting

• Fontconfig warning: “using without calling FcInit()”

  • This is generally harmless in GTK apps; GTK initializes Fontconfig internally. You can ignore it unless fonts fail to render.

• Indentation or Python errors

  • Ensure you’re running Python 3. Reinstall dependencies if needed.

• Cannot fetch models

  • Check Base URL, network connectivity, and API Key (if required).
  • Some local servers may not expose /v1/models exactly like OpenAI; ensure compatibility mode is enabled.

• “Unauthorized” or 401

  • Ensure your API key is correct and valid for the chosen endpoint.
  • The Authorization header is sent as Bearer {API_KEY} if provided.

• Dark mode not applied

  • Confirm GNOME Appearance is set to Dark.
  • Try: GTK_PREFER_DARK=1 python3 ollama_gui/code.py.

• Accent color not applied

  • Requires GNOME with accent color support and Adwaita-based theme.
  • The app uses the GNOME suggested-action class for accents; custom themes may not support accents.

Security

• API keys are stored in plaintext JSON at ~/.config/openai_gtk_client/settings.json for convenience. If this is a concern, prefer a secrets manager or environment variables in a modified build.

Roadmap / Ideas

• Streaming responses
• Markdown rendering and code block formatting tools
• Sidebar for chat history
• Export/import conversations
• About dialog and shortcuts help window
• GTK4 port with libadwaita widgets