jiavis

simple local voice assistant that runs entirely on your machine 24/7 with no cloud, api keys, or subscriptions (this was done a few weeks before clawdbot)

for me i ran phi-2 on lmstudio with a speaker microphone using whisper and python on an intel asus nuc 14 rvk

how it works

you speak → mic captures → whisper transcribes → phi-2 thinks → speaker responds

requirements

• windows 10/11
• python 3.8+
• microphone (tested with samson q2u)
• speaker (tested with jbl clip 5, yeah just what i had laying around)
• lm studio
• in my case my intel asus nuc 14 rvk

setup

1. install dependencies

pip install sounddevice numpy openai-whisper requests

2. download a model in lm studio

• open lm studio
• search for TheBloke/phi-2-GGUF
• download the Q4_K_M version

3. start the local server

• go to the developer tab in lm studio
• load the phi-2 model
• toggle the server on (runs on localhost:1234)

4. run jiavis

python jiavis.py

5. speak after you see "listening..."

architecture

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│  mic         │ ──▶ │     your pc  │ ──▶ │  speaker     │
│  (input)     │     │  (local ai)  │     │  (output)    │
└──────────────┘     └──────────────┘     └──────────────┘
                            │
          ┌─────────────────┼─────────────────┐
          ▼                 ▼                 ▼
   ┌────────────┐    ┌────────────┐    ┌────────────┐
   │  whisper   │    │  lm studio │    │  windows   │
   │  (stt)     │    │  + phi-2   │    │  tts       │
   └────────────┘    └────────────┘    └────────────┘

customization

edit the system prompt in jiavis.py to change the personality:

{"role": "system", "content": "you are jiavis, a helpful voice assistant..."}

why local?

• privacy: ur voice never leave machine
• free: no api costs or subscriptions
• fast: no network latency
• offline: works without internet

credits

built with whisper and lm studio

hope u enjoyed and bye