Back to Projects
Windows macOS Linux Cross-Platform (Windows, macOS, Linux)
FastAPI + Async Python 3.11+
Zero API Fees · Multi-LLM Broadcast · Chrome Extension Bridge

AI Spark — Local Multi-LLM Broadcast, Comparison & Synthesis Suite

AI Spark is an open-source, local multi-model AI desktop orchestrator and comparative web suite. It simultaneously broadcasts single prompts to ChatGPT, Gemini, and DeepSeek running securely in your browser, displays responses side-by-side in a comparative grid, and leverages Claude to synthesize a comprehensive, non-contradictory final answer with zero paid API costs.

AI Spark - Multi-Model AI Orchestrator
Architecture Showcase
Multi-Model Asynchronous Broadcast & Claude Neural Synthesis

Architectural & Engineering Highlights

Concurrent Multi-LLM Broadcast Engine

High-speed asynchronous asyncio pipeline broadcasting your single query simultaneously across ChatGPT, Gemini, and DeepSeek browser tabs with zero serial queuing latency.

Claude Neural Synthesis Core

Gathers generated responses from all three AI models in real time, parses AST Markdown and code blocks, and prompts Claude to formulate a structured, exhaustive, and unified answer without contradictions.

Dual Modes: Instant & Spark DeepThink

Switch seamlessly between Instant Mode (1-on-1 direct conversation with a single model for rapid prototyping) and Spark Mode (full parallel 3-way multi-model broadcast with automatic Claude synthesis).

Universal Multi-Browser CDP Driver

Native support for Brave, Chrome, Edge, Firefox, and Opera. Connects via Chrome DevTools Protocol (CDP :9222) directly to your default browser profile—preserving existing logins, cookies, and extensions.

Chrome Extension Bridge (Manifest V3)

Features a dedicated WebSocket bridge server (ws://127.0.0.1:41243) connecting an unpackable Chrome Extension for setups where direct profile locking is restricted, enabling zero-restart browser attachment.

100% Local Privacy & Multi-Modal Staging

Zero third-party cloud telemetry. Sessions are saved locally in SQLite (chat_history.db) with full search. Multi-modal files (PDFs, images, docs) are automatically distributed to all active model tabs.

Technical Comparison & Efficiency Matrix

How AI Spark compares to manual tab switching and commercial cloud API aggregators.

Feature & Capability AI Spark Engine Manual Multi-Tab Browsing Cloud API Aggregators
Recurring Subscription Cost $0.00 (Uses Web Sessions) $0.00 (High manual effort) $20 – $100+ / month
Simultaneous Query Broadcast Instant 1-Click Parallel Manual copy-paste to 4 tabs API rate-limited calls
Multi-Model Response Synthesis Automated Deep Claude Synthesis Manual mental cross-check Basic generic prompt summary
Data Privacy & Telemetry 100% Local FastAPI / Zero Logs Standard browser privacy Third-party server logging
Multi-Modal File Upload Distribution Automated CDP Multi-Tab Ingestion Manual file upload on each tab Strict token / file size limits
Local Chat History & Search Indexed SQLite Database (chat_history.db) Scattered across 4 web platforms Cloud stored history

Supported AI Ecosystem & Model Roles

OpenAI ChatGPT

Excels at logical reasoning, conversational nuance, coding boilerplate, and creative problem solving.

Role: Generalist & Reasoning

Google Gemini

Massive context window processing, live web grounding, research synthesis, and multi-modal understanding.

Role: Research & Grounding

DeepSeek

Deep chain-of-thought mathematical reasoning, competitive programming logic, and algorithm design.

Role: Logic & Code Verification

Anthropic Claude

Master synthesis engine reconciling contradictions, merging code solutions, and outputting clean Markdown reports.

Role: Chief Synthesizer

Cross-Platform Setup & Execution

Windows Windows 10 / 11

Initializes isolated .venv_windows, installs requirements, and launches Brave or Chrome in CDP mode:

run_windows.bat
Linux Linux (Ubuntu / Arch / Fedora)

Supports native, Snap, and Flatpak browser binaries with automatic XDG fallback profiles:

chmod +x run_linux.sh
./run_linux.sh
macOS macOS (Intel & Apple Silicon)

Works with system or Homebrew Python 3.11+, setting up .venv_macos automatically:

chmod +x run_macos.command
./run_macos.command

Ready to orchestrate multi-model AI workflows?

Download the latest release or explore the open-source repository on GitHub.