Lineworks

An desktop native LLM integrated text editor built for cowriting. Focuses on efficient and hassle-free usage.

*new* Discord: Arena Systems.

About

Designed for rapidly iterating on text. Generate at any position, quickly revert and regenerate, hotkeys for everything and stopping conditions useful for writing (Sentence, Line, Paragraph). Organized with multiple tabs and working areas that can be saved and restored to a file. Autosaves every two minutes and keeps the outputs from the current session in a searchable history, so you dont lose anything.

Local and remote generation is available (TogetherAI and OpenAI HTTP APIs are support). Different color schemes available (Light, Classic).

Usage

To use: Download Lineworks, extract the archive, and then run. NVIDIA or AMD* GPU acceleration is available to choose during install, otherwise choose CPU. Updating is done inside Lineworks via File->Update. Linux users run with bash source/start.sh.

Uses llama.cpp with its self-contained and quantized model format, GGUF. Many models are available, for starters try: Mistral-7B. .gguf models placed inside the models folder will be available to use in Lineworks.

Saving and loading is done with .json files, these store the current state of Lineworks. Autosaving is done every two minutes if a save file is open. Lineworks is intended to be operated via keyboard shortcuts, they are all displayed next to their corresponding action in the top menu bar.

*AMD on Windows is still experimental, see here for details.

Remote

Together.AI

Currently the easiest way to generate remotely since they are giving free credit to new accounts. First Sign up, then head to API Keys and copy your key. Switch the Lineworks backend from Local to Remote, put https://api.together.xyz/ as the Endpoint and your API Key as the Key. Press the connect button.

Websockets

Lineworks has its own websocket API which acts closer to Local than the HTTP APIs. Server hosting is done with python3.10 source/server.py, for example:

cd ~
git clone https://github.com/arenasys/lineworks
cd ~/lineworks/models
wget https://huggingface.co/TheBloke/Mistral-7B-v0.1-GGUF/resolve/main/mistral-7b-v0.1.Q5_K_M.gguf
cd ~/lineworks
pip install websockets==11.0.3 bson==0.5.10 cryptography==40.0.2 
pip install https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/wheels/llama_cpp_python-0.2.20+cu120-cp310-cp310-manylinux_2_31_x86_64.whl
python source/server.py --bind "127.0.0.1:8080"

Which will be accessible on ws://127.0.0.1:8080. Different llama-cpp wheels will be needed depending on the system: CUDA 12 (cu120), CUDA 11.8 (cu118), etc.