AI_Eli

An early version of ELI (Entropic Logical Interpreter), a local voice assistant for Linux that runs on Ollama. It listens for a wake word, works out what you asked, and either answers or does it on the PC: opening apps, folders and websites, media and volume, reading and writing files, making documents, and summarising PDFs.

Everything runs on your own machine: the model through your local Ollama, and speech recognition through Whisper (faster-whisper). The first time you speak to it, the Whisper model is downloaded once from Hugging Face.

Download

The releases page has a ready-to-run Linux AppImage. The rest of this page is for running it from source.

What you need

• Linux with PulseAudio or PipeWire
• Python 3.10 or newer
• Ollama with any chat model pulled
• A few system tools: espeak-ng, pulseaudio-utils (for paplay), libportaudio2, xdg-utils, playerctl, and wl-clipboard or xclip

On Ubuntu or Debian:

sudo apt install espeak-ng pulseaudio-utils libportaudio2 xdg-utils playerctl wl-clipboard

Setup

git clone https://github.com/ShadowESC95/AI_Eli.git
cd AI_Eli
python3 -m venv .venv
. .venv/bin/activate
pip install -r requirements.txt
ollama pull <any-model>

Then tell ELI which model to use. Put settings in ~/.config/AI_Eli/config as KEY=VALUE lines, or export them:

ELI_MODEL=<any-model>
ELI_USER_NAME=<your name>

Run

What	Command
Voice assistant	python Eli_PC_Control2_Working.py
List microphones	python Eli_PC_Control2_Working.py --list-devices, then set ELI_AUDIO_IN
Desktop window	python eli_pro_audio_gui_MKI.py
PDF and memory window	python eli_pro_audio_gui_MKII.py
Terminal chat	python -m eli_tools.eli_terminal

Say "Eli" or "hey Eli" (or "computer") before a request. Typing, key presses and pasting into other windows stay off until you say "activate full control", which lasts a minute. ELI_REQUIRE_ARM=0 turns that check off.

To summarise PDFs, put them in data/input and say "analyse my PDFs". The summary lands in artifacts/.

Optional

• Persona. scripts/build_persona_model.sh <any-model> builds an eli-persona model from modelfiles/eli-persona.Modelfile on top of the model you choose. Then set ELI_MODEL=eli-persona.
• Better voice. scripts/get_piper_voice.sh downloads a Piper voice into eli_tools/voices. Install Piper and set ELI_TTS_ENGINE=piper.
• Mic button in the desktop window needs PyAudio (sudo apt install portaudio19-dev && pip install pyaudio).
• Two machines with a shared clipboard. eli_tools/eli_listen.py turns speech into eli@<hostname>: ... on the clipboard, and eli_tools/eli_agent.py, running on each machine, carries out the ones meant for it. Set ELI_ALIAS_A and ELI_ALIAS_B to the two hostnames.

Settings

Put these in ~/.config/AI_Eli/config as KEY=VALUE lines, or export them. The AppImage reads only that file and variables named with AI_ELI_ in place of ELI_ (AI_ELI_MODEL=llama3.2); it ignores plain ELI_ variables, which other ELI versions use.

Setting	Default
ELI_MODEL	the newest chat model in Ollama	Ollama model for chat (also read from ELI_CHAT_MODEL or OLLAMA_MODEL)
ELI_ROUTER_MODEL	ELI_MODEL	a smaller model for working out commands
OLLAMA_BASE_URL	http://127.0.0.1:11434	where Ollama is
ELI_USER_NAME	none	what ELI calls you
ELI_WAKE	computer,eli,hey eli,okay eli	wake words
ELI_ALWAYS_AWAKE	0	1 answers without a wake word
ELI_AUDIO_IN	system default	microphone, by number or name
ELI_SR	the mic's own rate	microphone sample rate
ELI_OUTPUT_DEVICE	system default	PulseAudio sink to speak through
ELI_WHISPER	base	Whisper model size
ELI_WHISPER_DEVICE	cpu	cuda for an NVIDIA card
ELI_TTS_ENGINE	espeak	or piper
ELI_PIPER_MODEL	eli_tools/voices/en_US-amy-medium.onnx	Piper voice file
ELI_ARTIFACTS_DIR	./artifacts	where documents and summaries go
ELI_CONFIG	~/.config/AI_Eli/config	another settings file

What ELI remembers is kept in data/, artifacts/, state.json and memory.jsonl in the project folder (in ~/.local/share/AI_Eli for the AppImage), none of which go into git.

Licence

Apache 2.0, see LICENSE.