Palm1r/QodeAssist

QodeAssist – AI‑powered coding assistant for Qt Creator

What it is – QodeAssist is a Qt Creator plugin that brings large‑language‑model (LLM) capabilities directly into the IDE. It adds AI‑driven code completion, a chat panel, inline “quick‑refactor” edits, and a set of tools (file read/write, build, terminal, diagnostics, etc.) that the model can invoke. The plugin can work with a wide range of providers – local runtimes such as Ollama, llama.cpp, LM Studio, as well as cloud APIs like Anthropic Claude, OpenAI, Google Gemini, Mistral/Codestral, Qwen, DeepSeek, and any OpenAI‑compatible endpoint.

Key features

• Context‑aware code completion (single‑line and multiline) for C++ and QML, supporting both fill‑in‑the‑middle (FIM) models and chat‑style models.
• Chat assistant with side‑panel, bottom‑panel or detached window, history persistence, token‑usage display, and optional summarisation of long conversations.
• Quick Refactor – invoke the model to rewrite selected code inline, using a searchable library of custom instructions.
• Agent tools – the model can call functions to list/project files, read/write files, run builds, execute shell commands, fetch linter diagnostics, and manage TODO‑style tasks.
• Agent Skills – reusable folders of Markdown‑based instructions (compatible with the open Agent Skills spec) that can be loaded automatically, invoked via /skill, or set to “always‑on”.
• MCP (Model Context Protocol) server – expose QodeAssist’s tools over HTTP/SSE or stdio so other MCP‑compatible clients (Claude Code, VS Code, Cursor, etc.) can reuse the same project context.
• MCP client hub – connect to external MCP servers and make their tools available inside QodeAssist.
• Multi‑provider support – configure local or cloud models through a simple UI; quick‑setup presets cover the most popular services.
• Customizable prompt composition – per‑feature system prompts, open‑file context, and optional file‑attachment handling.

Installation

1. Extension‑registry method (recommended) – add the QodeAssist registry URL that matches your Qt Creator version (e.g. https://github.com/Palm1r/extension-registry/archive/refs/heads/qodeassist.tar.gz) via Qt Creator → Extensions → Browser. Install and update from the same screen.
2. Manual method – download the compiled plugin archive, place it in the Qt Creator plugins directory for your version (e.g. ~/Library/Application Support/QtProject/Qt Creator/plugins/16.0.0/ on macOS), then enable it via Help → About Plugins… → Install Plugin….

Configuration

• Use the Quick Setup dropdown to pick a preset (Claude, OpenAI, Mistral, Google Gemini, Qwen, DeepSeek) and enter the API key; the plugin fills in provider URL, model name, and prompt template automatically.
• For local models, follow the linked docs for Ollama, llama.cpp, or LM Studio to point QodeAssist at the appropriate endpoint.
• Advanced users can edit provider settings manually, enable/disable individual tools, and add global or project‑specific Agent Skills directories.

Usage

• Code completion – type normally; a hint appears after three characters (hint‑based mode) or completions are requested automatically (automatic mode). Press the configured key (default Space) to accept.
• Chat – open the chat panel, type queries, attach files/images, or invoke a skill with /skillname. The model can call tools like read_file or build_project during the conversation.
• Quick Refactor – select code, trigger the refactor command, and the model returns an edited snippet based on the chosen instruction template.
• MCP – enable the server in Tools → Options → QodeAssist → MCP Server to let other editors use the same project‑aware tools, or configure external MCP servers under MCP Client.

Project status

• The repository is archived. The author stopped development after a copyright‑notice dispute. The last released version remains downloadable, but no further updates or bug fixes are expected.

License – GPL v3.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.