ELI 0.1.1

ELI (Entropic Logical Interpreter), the first version: a local AI assistant for Linux with a desktop window, memory, voice and system control.

Get started

Download ELI-0.1.1-x86_64.AppImage, make it executable and open it:

chmod +x ELI-0.1.1-x86_64.AppImage
./ELI-0.1.1-x86_64.AppImage

That one file is everything ELI needs: Python and all its libraries, llama.cpp, the window, a language model (Qwen2.5 1.5B Instruct, Apache 2.0, from Qwen) and the Whisper speech model. Nothing leaves your machine.

To use another model, put any instruct model in GGUF format in ~/.local/share/eli-ai/models or beside the AppImage. ELI uses that one instead of the one it ships with.

Command
./ELI-0.1.1-x86_64.AppImage	the desktop window
./ELI-0.1.1-x86_64.AppImage --cli check	check the setup
./ELI-0.1.1-x86_64.AppImage --daemon	the background learning loop
./ELI-0.1.1-x86_64.AppImage --api	the local API on 127.0.0.1:8081

Its own folders and settings

ELI keeps everything in its own folders, apart from any other ELI version: memory, notes and documents in ~/.local/share/eli-ai, settings in ~/.config/eli-ai.

Settings go in ~/.config/eli-ai/settings as KEY=VALUE lines (the README lists them), for example ELI_N_GPU_LAYERS=-1. A variable named with ELI_AI_ in place of ELI_ works too: ELI_AI_N_GPU_LAYERS=-1 ./ELI-0.1.1-x86_64.AppImage. Plain ELI_ variables belong to other ELI versions and are ignored.

This build runs the model on the processor. For an NVIDIA card, install from source with a CUDA build of llama-cpp-python (see the README) and set ELI_N_GPU_LAYERS=-1.

Fixed since 0.1.0

• 0.1.0 needed the model as a second download; on its own it could not chat. The model is inside now.
• Closing the window crashed: ELI tried to save the conversation inside the AppImage, which is read-only. Everything ELI writes now goes to its own data folder.
• 0.1.0 used the same folders as later ELI versions (~/.local/share/eli, ~/.config/eli) and some shared files in the home folder. It has its own now.
• The speech model's onnxruntime queued usage telemetry for Microsoft. That is switched off.

Also here

• eli_ai-0.1.1-py3-none-any.whl and eli_ai-0.1.1.tar.gz: ELI as a Python package
• SHA256SUMS.txt: checksums of every file

Needs

• Linux on x86_64 with glibc 2.35 or newer (Ubuntu 22.04, Debian 12, Fedora 36 or later)
• About 2 GB of memory for the model
• espeak-ng for spoken replies; PulseAudio or PipeWire for sound
• If your system will not open AppImages, run it with --appimage-extract-and-run

Apache 2.0.