TabNeuron

Problem: "Which tab has that info?" — Scrolling through dozens of hidden tabs in browser bars, losing context.

• Solution: AI groups tabs on a 2D canvas by topic — visually find tabs by what they're about, not where they're buried.
• Canvas Editor: Drag and drop, rename, resize — full visual control.
• Mind Maps: Generate visual representations of tab groups to reveal hidden relationships.

Problem: Workspace and browser are disconnected — changes in one don't reflect in the other.

• Solution: Bidirectional sync with Chrome Tab Groups — canvas changes push to browser, browser changes pull to canvas.
• Workspace → Browser: AI suggests semantic groupings, you review and click "Apply" — browser tab groups update instantly.
• Browser → Workspace: Create groups or open tabs in browser — TabNeuron's polling API detects changes and pulls them onto the canvas.

Problem: Opening and reading across multiple tabs takes too long. Research means comparing, copying, synthesizing manually.

• Solution: Select multiple tabs and ask the AI to synthesize, compare, or extract data across all of them simultaneously.
• DOM Extraction: Page content and metadata fed directly into LLM context — no manual reading.
• Multi-Tab Chat: "Compare specs across 4 review tabs" → result saved locally.

Problem: Repetitive web research is manual — open tabs, extract data, compare, format, save — every time.

• Solution: Build no-code automated workflows where one agent's output becomes the next agent's input.
• Scraper Agent: Reads selected tabs, uses web_extract_links to find specs.
• Analyst Agent: Takes output, compares, formats a Markdown table.
• Output Agent: Writes executive summary, saves as persistent text node on canvas.

Problem: Can't tell what a tab contains without switching to it — title alone isn't enough.

• Solution: Automatic scaled screenshots, titles, URLs, and OG-tags extracted for every tab — instant visual context.
• Data Ownership: Save workspace states locally to disk. Close browser, research session remains safely backed up.

Problem: AI forgets who you are after every session. Long chats are expensive and slow.

• Solution: Auto-extracts facts, skills, preferences, and personas from conversations. Gets better over time.
• Auto-Compaction: Old conversations summarized → 93% token savings, full context preserved.
• Global Memory: Shared across all agents, import/export, fully manageable.

Problem: Repetitive file tasks are manual. Web research means opening multiple tabs, comparing, copying. Research takes forever.

• MCP Manager: Visual interface — enable/disable servers, configure settings, monitor health.
• File Operations: read, write, move, search, edit — all through conversation.
• Web Tools: Fetch, scrape, search, extract — AI has direct web access.
• External Servers: Connect Google Drive, GitHub, PostgreSQL, custom tools.

Problem: Data goes to cloud providers — no privacy. Or local models are hard to configure.

• Local: 🍋 Lemonade (NPU acceleration), Ollama, Llama.cpp, LM Studio — nothing leaves your machine.
• Cloud: OpenAI, Mistral, Gemini (free tier), and more — when you need power.
• Model Manager: Activate/deactivate models per task — 2-pane view (🍋 Lemonade only) for chat + embedding simultaneously.

• Portable: Runs from USB, extract anywhere, data stays under your control.
• Centralized Storage: All data in a single .tabneuron/ folder. Existing installs migrate automatically.
• Browser Bridge: Lightweight extension connects via local polling API — no data leaves your machine.