seerai

Powered by NanoGPT — Access 300+ AI models with flexible pay-as-you-go or subscription plans.

seerai is an intelligent research assistant plugin for Zotero 8 that integrates AI-powered chat, advanced search, and data extraction capabilities directly into your research workflow. Chat with your papers, extract structured data, and accelerate your literature review with a local-first, privacy-focused design.

---

Features

AI-Powered Chat Interface

• Contextual Conversations: Chat with AI about your selected papers with full context awareness.
• Smart Context Priority: Automatically prioritizes content sources:
  1. Zotero Notes (OCR note, and other notes for highest priority)
  2. Indexed PDF Text (Fast, efficient, however consumes a lot of tokens and may cause limit issues)
  3. OCR (Fallback for scanned documents with no indexed text)
• Multi-paper Support: Add multiple papers to a single conversation for comparative analysis.
• Streaming Responses: Real-time, token-by-token response rendering.
• Markdown & Math: Responses are formatted with syntax highlighting and LaTeX math support.
• Vision Support: Paste images directly into chat for multimodal analysis.
• Multimodal Generation: Generate images, videos, speech-to-text, and text-to-speech directly from chat.
• Attachments Upload: Add files to conversations via the context menu.
• Interactive Follow-ups: Deepen the conversation with streaming follow-up questions.
• Configurable Citations: Choose your preferred citation style for AI insights and chat.
• Smart Copy: Select and copy text with preserved formatting (Markdown) directly from chat bubbles.
• Enhanced Keybindings:
  • Enter: Insert new line
  • Shift+Enter: Send message
  • Ctrl+Shift+S: Toggle/Focus detachable window
• Detachable Window: Pop out the SeerAI interface into a standalone resizable floating window to maintain chat access while navigating your library.
• Themed UI: Enhanced dialogs and components with full theme support for a consistent look across Zotero's Light and Dark modes.
• Responsive Layout: Chat, tables, and search tabs dynamically adapt to panel width changes.

Semantic Search & Discovery

• RAG (Retrieval-Augmented Generation): Per-context embeddings with chunking, vector store, and semantic retrieval for large documents.
• Web Search: Integrated Firecrawl & Tavily support for finding high-quality full-text content.
• Semantic Scholar Agent: Advanced paper search with:
  • Advanced Filters: Fine-tune results by Year, Venue, and Citation Count.
  • AI Insights Config: Configure insight generation directly from the search panel.
• Smart Import:
  • PDF Discovery: Automatically finds and attaches PDFs during import.
  • Source Link: Fallback to source links if PDFs are unavailable.
  • Status Indicators: Clear feedback on import status (⬇️ Importing, ✅ Imported, ⚠️ Failed).
• Global Search Scope: Searching now extends across all libraries, including personal and group collections.
• Advanced Boolean Search: Robust support for nested logic (AND/OR/NOT), implicit phrasing, and markdown-aware matching.
• Smart Regex matching: Improved search precision with intelligent handling of word boundaries and special characters.

Agentic Chat & Tool Use

• Autonomous Agents: AI can now use tools to interact with your Zotero library and the web.
• Rich Tool Suite:
  • Search Tool: Search through your library with advanced filters.
  • Collection Tool: Manage collections and move items.
  • Note Tools: Read, create, and edit existing item notes for seamless research updates.
  • Tag Tool: Automatically generate and apply relevant tags to your research.
  • Read Tool: Extract text from PDFs and items for deep analysis.
  • Citation Tool: Generate citations and bibliographies.
  • Table Tool: Interact with and generate data for your Paper Tables.
  • Web Tool: Search the web and fetch content using Firecrawl or Tavily.
• Advanced Orchestration: Improved tool calling logic and iteration tracking for more robust agent performance.
• Model-as-a-Tool: Seamless integration with LLM tool-calling capabilities.

Papers Tables

• Structured Extraction: Extract specific data points from multiple papers into a comparative table.
• AI-Powered Columns: Define custom columns with AI prompts (e.g., "Methodology", "Sample Size").
• Inline Editing: innovative inline editor for column titles and prompts.
• One-Click Generation: Generate data for individual cells or entire columns instantly.
• Bulk Actions: Regenerate content or add selected papers to collections in bulk.
• Side Strip Actions: Unified controls for adding, removing columns, generating triggers, and settings.

OCR & Text Extraction

• Flexible OCR Options:
  • Mistral OCR: High-quality cloud OCR (Recommended).
  • DataLab.to: Reliable cloud-based extraction.
  • Local Marker: Run your own local OCR server for free, private processing.
• Auto-Processing: Automatically processes unindexed PDFs when needed.

Customizable AI

• Persistent API Keys: API keys are saved and persisted across all configured providers.
• Model Presets: Pre-configured settings for popular providers:
  • NanoGPT (300+ models, pay-as-you-go or subscription)
  • OpenAI (GPT-5, o3)
  • Anthropic (Claude Sonnet 4.5)
  • Google (Gemini 3 Pro)
  • DeepSeek, Mistral, Groq, OpenRouter
  • Local Models (Openai compatible endpoint, Ollama, LM Studio)
    • 12-16g Vram - Qwen3-4B-Thinking-2507
    • 24-32g Vram - gpt-oss-20b
    • 48-64g Vram - QwQ-32B
    • 96-128g Vram - Qwen3-Next-80B-A3B-Instruct
• Smart Rate Limiting: Per-model configuration for concurrency, RPM, and TPM to prevent provider errors.
• Per-Conversation Models: Switch models dynamically based on the task complexity.

---

Installation

From GitHub (Recommended)

1. Download the latest release (.xpi file) from Releases.
2. In Zotero, go to Tools → Add-ons.
3. Click the gear icon ⚙️ and select Install Add-on From File....
4. Select the downloaded .xpi file.
5. Restart Zotero.

From Source

# Clone the repository
git clone https://github.com/dralkh/seerai.git
cd seerai

# Install dependencies
npm install

# Build the plugin
npm run build

# The .xpi file will be generated in the root directory

---

Configuration

Go to Zotero → Settings → seerai to configure your AI providers and services.

1. AI Models

Use the Add Configuration button to set up your AI models.

• NanoGPT (Default): Pre-selected when adding a new model. Access 300+ AI models with flexible pricing — pay-as-you-go or subscription plans. Supports Apple Pay, Google Pay, credit card, and crypto. The model list is fetched automatically.
• Presets: Select from built-in presets (OpenAI, Anthropic, Ollama, etc.) for quick setup.
• Custom: Manually configure API URL, Key, and Model ID for any OpenAI-compatible provider.
• Default: Set a preferred model as your default for new conversations.

2. OCR Services

Choose your preferred text extraction engine:

• Mistral OCR: Requires Mistral API Key. Best for accuracy.
• Cloud (DataLab.to): Requires DataLab API Key.
• Local Marker Server: Requires running a local Python server.
  • URL: http://localhost:8001 (Default)
  • See Marker Project for setup.

3. Search Integrations

• Semantic Scholar: Add your API Key for higher rate limits and faster searches.
• Firecrawl: Add API Key to enable deep web search capabilities - local instance with (GitHub).
• Tavily: Add API Key for optimized search results tailored for AI agents.

4. MCP Server & API

Seer-AI now includes a Model Context Protocol (MCP) server and a local API for external integrations.

• MCP Server: Located in mcp-server/. Allows external LLMs (like Claude Desktop) to interact with your Zotero library. See the MCP Setup Guide for configuration instructions.
• Local API: Provides endpoints for chat, tool execution, and library management.
  • Settings → seerai → API.
  • Default Port: 23119

This mode requires sophisticated models with strong tool/function-calling capabilities to function properly.

---

Usage Guide

Chatting with Papers

1. Select a paper (or multiple) in your library.
2. Open the SeerAI sidebar tab.
3. (Optional) Customize context inclusions via the settings icon (Abstracts, Notes).
4. Type your question or use templates from the Prompt Library (Book icon).

Detachable Window

• Pop-out: Click the ⇱ button in the SeerAI tab bar to open a floating window.
• Hotkey: Press Ctrl+Shift+S to instantly detach, toggle, or focus the window.
• Auto-Sync: The detached window automatically updates its context when you select different items in Zotero.
• Attach: Close the floating window or click the dock button (within the sidebar placeholder) to return to the sidebar.

Creating Data Tables

1. Open the Tables tab in the main view.
2. Click + on the side strip to add a new column.
3. Define the column header and AI prompt (e.g., "What is the sample size?").
4. Drag and drop papers into the table.
5. Click Generate on cells to extract data.

Prompt Library

• Access via the Book Icon 📖 in chat.
• Use built-in templates (Summarize, Critique, Compare).
• Create custom templates with placeholders:
  • !: Saved Prompts
  • /: Papers
  • ^: Folders
  • ~: Tags
  • @: Authors
  • #: Topics

---

Future Implementations Ideas

Enhanced search functionality to help users find relevant literature more effectively.

• Autocomplete: Intelligent suggestions for tags, creators, and collections as you type.
• Complex Queries: Support for boolean logic (AND/OR) and nested search conditions (e.g., "Title contains X AND Year > 2020").
• Field-Specific Search: Dedicated filters for titles, authors, years, and tags.
• Citation References: Inline citations within tables and chat during generation.
• Internal MCP Presets: Custom support for MCP JSON presets for streamlined integrations.
• Connectors: External service connectors for extended workflows.

---

Development

Prerequisites

• Node.js 18+
• Zotero 8

Project Structure

The codebase follows a modular architecture:

seerai/
├── addon/                 # Zotero integration files (XUL/XHTML)
├── src/
│   ├── modules/           # Core feature modules
│   │   ├── chat/          # Chat engine & state
│   │   │   └── rag/       # RAG pipeline (chunker, embeddings, retrieval, vector store)
│   │   ├── assistant.ts   # Main assistant logic
│   │   ├── firecrawl.ts   # Firecrawl integration
│   │   ├── nanogptWeb.ts  # NanoGPT web integration
│   │   ├── ocr.ts         # OCR implementation
│   │   ├── openai.ts      # LLM & multimodal client
│   │   ├── semanticScholar.ts # Semantic Scholar integration
│   │   └── preferenceScript.ts # Preferences logic
│   ├── utils/             # Utility functions
│   └── hooks.ts           # Zotero event listeners
└── package.json

Commands

npm start       # Start dev server with hot reload
npm run build   # Build for production
npm run lint:fix # Fix code style issues

---

Contributing

Contributions are welcome!

1. Fork the repo.
2. Create a feature branch (git checkout -b feature/MyFeature).
3. Commit changes (git commit -m 'Add MyFeature').
4. Push to branch (git push origin feature/MyFeature).
5. Open a Pull Request.

---

License

MIT License - see LICENSE for details.

Acknowledgments

• NanoGPT — Partner
• Zotero
• Mistral AI
• Semantic Scholar
• Firecrawl
• Marker