LlamaLink

A sleek GUI frontend for llama.cpp
Search, download, and chat with local LLMs in one app.

_{If this project helps you, a coffee helps me keep working on it.}

---

Features

Model Management

• Browse and download GGUF models directly from HuggingFace
• Sort by downloads, likes, date, or trending
• View available quantizations (Q4_K_M, Q8_0, IQ3_S, etc.) with file sizes
• Download with progress bar, speed display, ETA, and resume support
• Recursive model folder scanning with automatic detection
• Confirmation-gated model pruning with protection for active and saved model references

Server Control

• Launch and manage llama-server with full parameter control
• Or connect to an already-running server (any OpenAI-compatible endpoint)
• Auto-detect llama-server from PATH and common install locations
• Context size, GPU layers, threads, flash attention, mlock toggles
• Embedded server log viewer

Chat Interface

• Streaming responses with live token-by-token display
• Markdown rendering: code blocks, inline code, bold, italic
• Tokens/sec speed display during and after generation
• System prompt support
• Parameter presets: Default, Creative, Precise, Code, Roleplay
• Adjustable temperature, top_p, top_k, repeat penalty, max tokens
• Prompt inspection, grammar controls, token probabilities, and configurable energy estimates
• Local LoRA fine-tune kickoff through llama.cpp's finetune example

Chat History

• Auto-saves conversations locally
• Load, export (Markdown / JSON / Text), and delete past chats

Design

• Catppuccin Mocha dark theme throughout
• Responsive split-panel layout
• Window position and all settings persist between sessions

Installation

Portable EXE (Recommended)

Download LlamaLink.exe from Releases and run it. No installation required.

Submission-ready package-manager manifests are included under packaging/winget/ and packaging/chocolatey/.

From Source

git clone https://github.com/SysAdminDoc/LlamaLink.git
cd LlamaLink
python llamalink.py

Dependencies (PyQt6, requests) are auto-installed on first run.

Quick Start

1. Download a model - Go to the "Download Models" tab, search for a model (e.g. llama, qwen, mistral), pick a quant, and download it
2. Set server path - Browse to your llama-server.exe (auto-detected if on PATH)
3. Select model - Your downloaded model appears automatically in the dropdown
4. Start server - Click "Start Server" and wait for the "Running" indicator
5. Chat - Switch to the Chat tab and start talking

Connecting to an Existing Server

Uncheck "Launch server", enter the URL (e.g. http://127.0.0.1:8080), and click Connect. Works with any OpenAI-compatible API endpoint.

Requirements

• llama.cpp - Download from llama.cpp releases
• Python 3.8+ (if running from source)
• NVIDIA GPU recommended (auto-detected, CPU-only works too)

HuggingFace Token

Public models work without authentication. For gated/private models, set the HF_TOKEN environment variable:

set HF_TOKEN=hf_your_token_here
python llamalink.py

Building from Source

pip install pyinstaller
pyinstaller llamalink.spec

The spec produces a one-file dist/LlamaLink.exe, wires the Qt runtime hook, calls freeze_support(), and embeds the application icon. The current C#/.NET release is built separately with dotnet publish as described above.

To produce the self-contained ARM64 Windows build, run:

dotnet publish src/LlamaLink.csproj -c Release -r win-arm64 -o dist-arm64 --self-contained true

License

MIT