ChatQT

A native AI chat client for the KDE Plasma desktop

Chat with AI models through multiple providers — directly from your desktop

---

About

ChatQT is a native KDE Plasma application for chatting with AI models. Built with Qt6/QML and Kirigami, it integrates seamlessly into the KDE desktop with a system tray icon, native look and feel, and Plasma-style dialogs.

Features

Providers

• Ollama — Connect to local models running via Ollama with automatic model discovery
• OpenAI Compatible — Connect to any OpenAI-compatible API (OpenAI, DeepSeek, Groq, etc.) with multi-provider support and connection testing
• OpenClaw — Connect to OpenClaw instances with multi-instance support (experimental)
• OpenCode — Built-in OpenCode server management with start/stop/restart controls and auto-start (experimental)
• Pi — Built-in Pi process management via RPC mode with auto-detect and auto-start (experimental)

Chat

• Streaming responses — Real-time token-by-token response display
• Thinking mode — Toggle extended thinking/reasoning for supported models
• Session management — Persistent sessions with sidebar, auto-restore on launch
• Cancel and stop — Cancel pending requests or stop mid-stream
• Auto-scroll — Automatic scrolling with manual override option

MCP (Model Context Protocol)

• Remote MCP servers — Connect to Streamable HTTP MCP servers
• Local MCP servers — Run stdio-based MCP servers as subprocesses (JSON-RPC)
• Built-in servers — Pre-configured Bash MCP and Filesystem MCP servers
• Tool calling — Automatic tool call detection, execution, and follow-up with configurable depth limit
• Server status — Real-time connection status and available tool count

Skills and Agent

• Skill discovery — Scan folders for SKILL.md files and inject them into chat context as system prompts
• Agent instructions — Load an AGENTS.md or CLAUDE.md file to provide persistent AI instructions
• System prompt builder — Automatically combines skills and agent instructions into a system message

Desktop Integration

• System tray — Minimizes to system tray, click to toggle visibility
• Flatpak support — Available as a Flatpak with proper sandboxing utilities

Screenshots

Main Window

Settings — Agent

Settings — OpenAI Compatible

Settings — Skills and MCP Servers

Installation

Building from Source

For build instructions, see BUILD.md.

Flatpak

A Flatpak manifest is available at org.koderoots.chatqt.json.

Experimental Features

ChatQT includes experimental features that are disabled by default. These features are functional but may have rough edges, change between releases, or lack full polish.

Enabling Experimental Features

Set the environment variable before launching:

CHATQT_ENABLE_EXPERIMENTAL_FEATURES=1 chatqt

Or for development builds:

CHATQT_ENABLE_EXPERIMENTAL_FEATURES=1 ./build/bin/chatqt

Once enabled, new provider options appear in Settings → General.

Experimental Providers

Provider	Description
OpenClaw	Connect to OpenClaw agent instances. Supports multiple instances with URL/token configuration and connection testing. Requires the OpenAI-compatible Chat Completions endpoint enabled in OpenClaw.
OpenCode	Manages an OpenCode server process directly from ChatQT. Configure the binary path, auto-detect, start/stop/restart, set host/port, and view server logs. Supports auto-start on launch and auto-restart on crash (up to 3 attempts).
Pi	Connects to a Pi coding agent via RPC mode (stdin/stdout JSONL protocol). Configure the binary path, auto-detect, start/stop, and view logs. Supports auto-start when Pi is the active provider.

License

This project is licensed under the GPL-3.0 License - see the LICENSE file for details.