Home Features How It Works Models Install GitHub
Voice-Driven · Local First · Open-Source
 ____         ___        _              _          ____  
/ ___|       / _ \      | |            / \        |  _ \ 
\___ \      | | | |     | |           / _ \       | |_) |
 ___) |  _  | |_| |  _  | |___   _   / ___ \   _  |  _ < 
|____/  (_)  \___/  (_) |_____| (_) /_/   \_\ (_) |_| \_\

Systematic Online or Local Autonomous Robot

Say its name. It listens. SOLAR is a voice-activated AI agent that runs fully on your Windows machine — handling calls, sending messages, writing code, and controlling your desktop through natural speech. No cloud required.

Scroll

// voice commands

// capabilities

What SOLAR Can Do

Every feature below is implemented in real, running Python — not a roadmap.

📞
Autonomous WhatsApp Call Handler
SOLAR monitors your desktop for incoming WhatsApp calls using Win32 EnumWindows. It screenshots the notification, runs it through a vision model to read the caller's name, then asks you by voice whether to answer — or declines with a custom voicemail, all autonomously.
SOLAR:"Incoming call from Anna Müller. Should I answer?"
You:"No, I'm busy." → SOLAR answers, leaves VM, hangs up
Vision AI · Win32
👂
Custom Wake Word
Define your agent's name in config.yaml. SOLAR listens 24/7 using fully offline Vosk STT — no internet, no cloud microphone, zero latency.
Always On · Offline
💬
Cross-Platform Messaging
Send messages on WhatsApp, Discord, and Instagram using image recognition to navigate each app's UI. No official API required.
You:say happy birthday to Thomas on whatsapp
WhatsApp · Discord · Instagram
💻
Voice Code Generation
Say "code a…" and SOLAR generates it with Malicus7862 DeepSeek Coder, auto-copies to clipboard, then describes what was built.
You:code a python web scraper
Malicus7862 Malicus7862 DeepSeek Coder 6.7b
🖥️
App & Window Manager
Open any app or URL and move its window to any connected monitor. Multi-monitor aware, including negative-coordinate secondary screens.
You:open youtube.com
Multi-Monitor · Win32
📅
Voice Calendar Events
Describe a meeting in plain speech. SOLAR parses the time with dateparser, creates an .ics file, and opens it straight in your calendar app.
You:deadline from 2pm to 4pm name it sprint review
Calendar · .ics
🧩
Multi-Model AI Routing
SOLAR automatically routes each task to the right model via Ollama on localhost:11434. Vision tasks go to the multimodal model, code to Malicus7862 DeepSeek Coder, writing to DeepSeek v2, and general chat to the main model. No API keys, no data leaving your machine.
llama3.2-vision:11b Malicus7862/deepseekcoder-6.7b-jarvis-gguf:latest deepseek-v2:16b Vosk STT Kokoro TTS
Fully Local · Ollama

// runtime flow

How SOLAR Works

Every interaction starts with your voice. SOLAR routes it through local models — no cloud required.

🎙️
Wake Word
Vosk listens 24/7 for your custom name from config.yaml
🔊
Hear Command
A beep confirms activation. Full command transcribed offline
🧠
Route & Execute
Intent matched — messaging, coding, calls, apps, or chat
🔈
Speak Back
Kokoro TTS voices the response in your configured name & gender

// ai stack

The Model Stack

Pick fully local Ollama models or Ollama's cloud models — STT and TTS are always local either way.

Local Models localhost · no internet
Model
Type
Used For
llama3.2-vision:11b
Vision
Call ID · general chat
Malicus7862/deepseekcoder-6.7b-jarvis-gguf:latest
Code
Voice code generation
deepseek-v2:16b
Text
Email · writing tasks
Vosk small-en-us
STT
Wake word · transcription
Kokoro TTS
TTS
Voice responses
Ollama Cloud Models localhost:11434 · Ollama cloud
Model
Type
Used For
Creative-Crafter/SOLAR-gemma3:27b_cloud
Vision
Call ID · general chat
qwen3-coder-next:cloud
Code
Voice code generation
deepseek-v3.2:cloud
Text
Email · writing tasks
Vosk small-en-us
STT
Same as local
Kokoro TTS
TTS
Same as local
Powered by ollama.com cloud GPUs · Browse all →

// quick start

Choose Your Setup

Pick the install that fits your hardware. Both handle dependencies and config automatically.

Local Install
All models run on your GPU · No cloud inference
~20 GB VRAM
PowerShell — Administrator
PS C:\> irm https://solar.creative-crafter.de/install-local.ps1 | iex
llama3.2-vision:11b Malicus7862/deepseekcoder-6.7b-jarvis-gguf:latest deepseek-v2:16b
Cloud Install
Ollama cloud models · No powerful GPU needed
Ollama account
PowerShell — Administrator
PS C:\> irm https://solar.creative-crafter.de/install-cloud.ps1 | iex
Creative-Crafter/SOLAR-gemma3:27b_cloud qwen3-coder-next:cloud deepseek-v3.2:cloud

🛡️  Both scripts require Administrator PowerShell  ·  Vosk STT & Kokoro TTS are shared between both setups

🪟 Windows 10 / 11
🐍 Python 3.11.0
🦙 Ollama v0.12+
🎙️ Microphone
💾 ~20 GB free (local models)

Open Source, Always

Every line of SOLAR is public. Fork it, extend it, add new skills in skills.py, or swap in different Ollama models. The project is built to be hacked on.

Apache 2.0License
PythonLanguage
v0.1Alpha
View on GitHub