quill-code

A native desktop home for Quill, a coding agent that runs on your own models. Point it at a llama.cpp server (or another OpenAI-compatible one), pick the model that powers Quill, and hand it work: it reads and edits your projects, runs commands in your system's own shell, shows every diff, and takes the git side all the way to a pull request.

Principles

• Native. Rust and GPUI, drawn on the GPU. No browser engine renders the interface.
• No telemetry, ever. The app talks only to the model servers you configure, to your git remotes, and to GitHub when you publish a repository or open a pull request.
• MIT licensed, with permissively licensed dependencies only.
• Built for local models. Tool calls, context windows and thinking controls are detected from the server rather than configured by hand.

Status

Early development, Windows first. Working today:

• Projects and threads, with several agents running at once and a sidebar that shows each thread's branch, commit, uncommitted work, unpushed commits, pull request and model.
• A tool-calling agent that lists, reads, searches, writes and edits files, runs commands, commits and opens pull requests, under Ask first, Trust edits and Autopilot access modes.
• Isolated copies (git worktrees) so parallel threads never collide, a commit dialog that can push, open a pull request page or publish a new GitHub repository, and new projects that can start on GitHub with a first branch of your choosing.
• Slash commands, @ file mentions, image and file attachments with previews, pinned models with Ctrl+1 to 9, a context ring, a PowerShell drawer and saved history.

GitHub sign-in goes through Git Credential Manager, so quill-code never stores a token.

Building

Every platform needs:

• Rust 1.88 or newer, installed with rustup. The default stable toolchain is enough.
• Git on the PATH, for cloning, commits, branches and isolated copies.
• A model server to talk to, such as llama.cpp's llama-server, or any server with an OpenAI-compatible /v1 API.

Then, from the repository folder:

cargo run -p quill-code --release

Windows

Windows is the platform quill-code is developed and tested on.

1. Install the Visual Studio Build Tools with the Desktop development with C++ workload, which brings the MSVC compiler and the Windows SDK.
2. Install Rust with rustup-init.exe, keeping the default x86_64-pc-windows-msvc toolchain.
3. Install Git for Windows. It includes Git Credential Manager, which handles GitHub sign-in.

macOS

1. Install Xcode from the App Store, not only the Command Line Tools: the interface's Metal shaders are compiled at build time with Xcode's metal tool. Open Xcode once to accept its licence, then run sudo xcode-select -s /Applications/Xcode.app. If the build says the metal tool is missing, run xcodebuild -downloadComponent MetalToolchain.
2. Install Rust with rustup.
3. Git comes with Xcode. For GitHub sign-in, install Git Credential Manager or any credential helper that stores a GitHub token.

Linux

Debian and Ubuntu:

sudo apt install build-essential pkg-config git libssl-dev libfontconfig-dev libxkbcommon-dev libxkbcommon-x11-dev libwayland-dev libvulkan1 mesa-vulkan-drivers

Fedora:

sudo dnf install gcc gcc-c++ pkgconf-pkg-config git openssl-devel fontconfig-devel libxkbcommon-devel libxkbcommon-x11-devel wayland-devel vulkan-loader mesa-vulkan-drivers

The interface draws with Vulkan, so the machine needs a working Vulkan driver; the Mesa drivers above cover most GPUs. For GitHub sign-in, install Git Credential Manager or any credential helper that stores a GitHub token.

macOS and Linux builds go through the same code but have not been tested yet.

Checks

cargo test --workspace
cargo clippy --workspace --all-targets
cargo deny check

cargo deny is installed with cargo install cargo-deny.

Licence

MIT, see LICENSE. Bundled Open Sans is under the SIL Open Font License, see assets/fonts/OFL.txt. The llama.cpp mark is from the llama.cpp project (MIT, Copyright (c) 2023-2026 The ggml authors).