LocalLLM

LocalLLM is a lightweight Tauri desktop control panel for running llama.cpp locally. It helps you discover GGUF models, configure llama-server, preview the launch command, save per-model presets, and open the local Web UI.

App Preview

Features

• Discover llama-server, llama-cli, hf, and huggingface-cli from PATH.
• Install supported prebuilt llama.cpp Windows and Ubuntu Linux release assets from inside the app.
• Save executable paths, model directory, manual GGUF entries, and server defaults.
• Scan a model directory recursively for .gguf files.
• Add manually selected .gguf files outside the model directory.
• Download GGUF files from Hugging Face through hf download.
• Preview the exact llama-server command before launch.
• Explain the active command in a compact helper below the preview.
• Estimate VRAM from GGUF metadata when available.
• Configure runtime, KV cache, GPU/memory, sampling, speculative decoding, chat/reasoning, multimodal, Jinja, tools, embeddings, and verbosity options.
• Save per-model presets and automatically return to the selected model's own preset when changing models.
• Start, stop, and open the local llama.cpp server.
• Run llama-server hidden in the background with log capture, or in a visible terminal.
• Warn before starting when another background llama-server process is already running.
• Reset LocalLLM's saved settings/cache without deleting downloaded models.
• Create, rename, duplicate, import, export, reset, default, and safely delete versioned Agent profiles with per-profile workspace, instructions, goal, reasoning, permissions, and execution limits.
• Recover the previous valid configuration automatically when a staged config write is interrupted or the newest file is malformed.

Runtime Requirements

The built app does not require Node.js or Python to run.

You only need the tools you want LocalLLM to control:

• llama-server is required to serve models.
• llama-cli is optional.
• hf or huggingface-cli is optional for Hugging Face downloads. LocalLLM can use curl for a single selected file, but glob patterns such as *.gguf need the Hugging Face CLI.
• GGUF model files.

On Windows and Ubuntu-based Linux distributions, the app installs prebuilt official llama.cpp release assets from the Settings panel. Ubuntu Linux installs use the official *-bin-ubuntu-*.tar.gz packages and require curl and tar on PATH; LocalLLM does not compile llama.cpp from source. You can also browse to your own llama-server and llama-cli binaries.

How To Use

1. Open LocalLLM.
2. In Settings, set or install llama.cpp.
3. Set the model directory where your .gguf files live.
4. Press Rescan to load models.
5. Pick a model from the model dropdown or model list.
6. Adjust runtime options and presets.
7. Press Start.
8. Press WebUI or switch to the Web UI tab after the server starts.

Useful controls:

• New Preset resets the current model preset to a clean default configuration.
• Profile editor lets you create, save, select, and delete presets.
• Tools all, Jinja, Embeddings, and Verbose are quick toggles near Extra args.
• Extra args is appended to the generated llama-server command.
• Reset everything in Settings clears LocalLLM settings, presets, cache, logs, theme, and layout. It does not delete GGUF files or your llama.cpp install.

Development Setup

Install:

• Node.js 22.19 or newer (required by the bundled Pi development dependency)
• Rust stable
• Tauri prerequisites for your OS

Install JavaScript dependencies:

npm install

Run in development mode:

npm run tauri dev

Run the frontend-only build:

npm run build

Run the current UI regression test:

npm test

Profile format, migration behavior, dependency decisions, and compatibility notes are documented in docs/profile-and-dependency-migration.md.

Build Release

Build for the current operating system:

npm run tauri build

Windows bundles are written to:

• src-tauri/target/release/localllm.exe
• src-tauri/target/release/bundle/nsis/
• src-tauri/target/release/bundle/msi/

Linux and macOS builds should be produced on their native operating systems or with CI runners for those systems. Tauri desktop bundles are OS-specific.

CI Release Builds

GitHub Actions organizes release builds by operating system (SO), splitting processor architectures (x64 and arm64) within each workflow:

• Linux (x64 & arm64): .github/workflows/build-linux.yml
• Windows (x64 & arm64): .github/workflows/build-windows.yml
• macOS (x64 & arm64): .github/workflows/build-macos.yml

Tag, release, and manual release uploads prefix uploaded assets with the workflow platform, such as windows-x64- or macos-arm64-, so artifacts from different architectures stay separate.

Linux Build Notes

Install Node.js, Rust, and Tauri's Linux system dependencies. On Ubuntu/Debian, the dependency list depends on your distro version, but commonly includes WebKitGTK, GTK, AppIndicator, librsvg, curl, wget, and build tools.

Then run:

npm install
npm run tauri build

Linux artifacts are produced under src-tauri/target/release/bundle/.

macOS Build Notes

Build on macOS with Xcode Command Line Tools installed:

npm install
npm run tauri build

macOS artifacts are produced under src-tauri/target/release/bundle/.

Repository Notes

Generated folders such as node_modules, dist, and src-tauri/target are ignored and should not be committed.