INTRAFIND Logo iHub

30+ ready-to-use AI apps for your team. Self-hosted. No prompting expertise needed.

License: MIT Latest Release Docker Image

iHub Apps is a full-stack, open-source AI platform that gives your team instant access to 30+ pre-built AI applications — no prompting expertise, no complex setup, and no data leaving your control. Connect any LLM (OpenAI, Anthropic, Google, Mistral, or local models), integrate your organization's knowledge sources, and deploy securely on-premise or in the cloud.

Unleash creativity and productivity in your organization with iHub — the platform that brings the power of generative AI directly into your secure corporate environment. Whether you need to compose emails, generate and edit texts, translate content, analyze files, or brainstorm ideas, iHub offers a suite of intuitive micro-apps designed for everyday business needs. With no prompting expertise required and full control over your data, iHub lets you work creatively and securely — on-premise or with carefully selected cloud models. Seamlessly integrate iHub with your existing IntraFind solutions and experience a unified platform for search, knowledge-based answers, and creative AI applications — all free and open source.

INTRAFIND iHub Startpage

iHub is developed by IntraFind Software AG — made with ❤️ from our teams in Berlin, Bonn, Munich + Remote

The Software is free-of-use and "AS-IS" without warranty of any kind. — License Details

For enterprise-grade support, custom features, or professional services, contact us at sales@intrafind.com.

No Node.js, no Docker, no dependencies. Download the standalone binary and you're running:

The simplest way to install — a single command handles everything:

curl -fsSL https://raw.githubusercontent.com/intrafind/ihub-apps/main/install.sh | sh

OR

👉 Download the latest release

Restricted environment? Use the .base64.txt files: base64 -d ihub-apps-v*-linux.tar.gz.base64.txt > ihub-apps.tar.gz

# Linux / macOS
tar -xzf ihub-apps-v*-linux.tar.gz
cd ihub-apps-v*
./ihub-apps-v*-linux

:: Windows — extract the .zip, then:
ihub-apps-v*-win.bat

Open http://localhost:3000 → Go to Settings → Models → Add your API key

🎉 Done! iHub auto-configures everything on first run. No .env file, no database, no manual setup.

Other install methods: One-Line Installer · Docker · npm (for developers)

iHub ships with a comprehensive library of ready-to-use AI apps covering the most common business workflows:

All apps are fully configurable via the admin interface — customize prompts, connect knowledge sources, set model preferences, and control access per user group. New apps can be created without coding.

iHub is built from a handful of building blocks. Knowing which one to reach for is the fastest way to turn an idea into a working use case:

Providers and models. Providers hold the technical connection — endpoint and credentials — to an external or local service; API keys are administered centrally and stored encrypted at rest. Models (GPT, Claude, Gemini, Mistral, or a local model) are what apps and users select. It pays to split models by use case: internal or sensitive scenarios on a local or dedicated model, public ones on a cloud model. Model choice can be narrowed per app (allowedModels) or fully predetermined (disallowModelSelection).

Apps vs. prompt library vs. skills. Apps are standardized use cases with a fixed frame — prompt, variables, tools, sources, model — and are rolled out per user group, including groups mapped from LDAP or OIDC. The prompt library is the flexible counterpart inside general chat, but a flat list of many prompts becomes hard to navigate. Skills package complex, multi-step procedures and reusable rules modularly: rather than growing one giant app prompt, split a workflow such as a story generator into separate skills for interviewing, structuring, and drafting.

Sources. Sources give the model its knowledge: local files (Markdown, PDF, text, JSON), URLs, iFinder documents, and internal pages. URLs suit content that changes regularly, at the risk that page boilerplate adds noise to the context; local files give a more controlled context but must be updated by hand. Each source is either loaded into the prompt up front or fetched on demand as a tool.

Tools. Tools extend iHub beyond text generation — web search, API calls, screenshots, people search, Jira, iFinder, and further systems via MCP.

In short: apps frame the use case, skills define reusable procedures, sources deliver the knowledge, and tools let the model act. Groups decide who gets what.

📖 Full Core Concepts Guide

Your data never leaves your infrastructure. Deploy on-premise, in your private cloud, or air-gapped. Connect to local LLMs (LM Studio, Jan.ai, vLLM) for complete privacy.

One interface for OpenAI GPT-4o, Anthropic Claude, Google Gemini, Mistral, and any OpenAI-compatible model. Switch models per app or let users choose. No vendor lock-in.

Connect your organization's knowledge: local files, SharePoint, enterprise document systems (iFinder), web pages, and databases. AI answers grounded in your content.

Every app ships with expert-crafted prompts. Your team gets instant value — no AI expertise, no prompt engineering, no training required.

Multi-mode authentication: Anonymous, Local, OIDC (Google, Microsoft, custom), and proxy auth. Group-based permissions. Hierarchical access control. CORS support for embedded deployments.

Standalone binary, Docker, npm, or PWA. Auto-configuration on first run. Multi-worker clustering for production. No database required.

Clean React SPA with dark/light mode, mobile-friendly design, real-time streaming responses, and full internationalization (English, German, and more).

Add new apps, models, and knowledge sources through the admin UI. JSON-based configuration. REST API for integration. Full source code available for deeper customization.

The simplest way to install — a single command handles everything:

curl -fsSL https://raw.githubusercontent.com/intrafind/ihub-apps/main/install.sh | sh

# Install and start immediately:
curl -fsSL https://raw.githubusercontent.com/intrafind/ihub-apps/main/install.sh | sh -s -- --start

CLI Options:

Environment Variables:

Post-install steps:

1. Edit ~/.config/ihub-apps/.env to add your API keys
2. Run ihub-apps to start the server
3. Open http://localhost:3000

Windows users: Download the .zip from GitHub Releases — the shell installer does not support Windows.

• ✅ Detects OS and architecture automatically
• ✅ Offers Docker if available on your system
• ✅ Verifies download integrity with checksums
• ✅ Generates a secure JWT secret automatically
• ✅ Upgrade-safe — re-run at any time to update

The fastest way to run iHub — a single executable with zero dependencies.

# Download, extract, run
tar -xzf ihub-apps-v*-linux.tar.gz && cd ihub-apps-v* && ./ihub-apps-v*-linux

• ✅ Zero dependencies — no Node.js, no Docker required
• ✅ Auto-setup — creates default configuration on first run
• ✅ Cross-platform — Windows, macOS, Linux binaries
• ✅ Production-ready — optimized single executable
• ✅ Built-in updates — update via Admin UI or --update CLI flag

📥 Download from GitHub Releases

docker run -d \
 -p 3000:3000 \
 -v $(pwd)/contents:/app/contents \
 --name ihub-apps \
 ghcr.io/intrafind/ihub-apps:latest

• ✅ Automatic local contents mounting
• ✅ Multi-platform support (Linux, macOS, Windows)
• ✅ Production-hardened container image

📖 Full Docker Guide · Docker Quick Reference

git clone https://github.com/intrafind/ihub-apps.git
cd ihub-apps
npm run setup:dev
npm run dev
# Open http://localhost:3000

Best for: customization, contributing, building new apps.

📖 Developer Setup Guide

• Multi-provider: OpenAI, Anthropic Claude, Google Gemini, Mistral — unified API
• Central provider management: Endpoints and API keys administered in one place, encrypted at rest
• Local LLMs: LM Studio, Jan.ai, vLLM — complete privacy, zero API costs
• Per-app model control: Restrict the selectable models or fix the model entirely
• Streaming responses: Real-time token streaming via Server-Sent Events
• Structured output: JSON schema validation for AI responses
• Tool calling: Function calling and agentic workflows
• Vision: Image understanding on models that support it
• Thinking models: Extended reasoning support (Claude, o1-series)
• Compare Mode: Send one input to two models and compare the answers side by side

• Filesystem: Local markdown, text, and JSON files as AI context
• Web pages: Intelligent content extraction from any URL
• Enterprise docs: iFinder document management integration
• Internal pages: Reuse FAQ and help pages as AI context
• Multi-source: Combine multiple knowledge sources per app
• On-demand or up front: Load a source into the prompt or let the model fetch it as a tool
• Admin interface: Create, test, and preview sources without coding

• Apps: Standardized use cases with a fixed prompt, variables, tools, sources, and model
• Skills: Reusable, modular instruction packages (SKILL.md + references) for multi-step procedures
• Prompt library: Searchable, reusable prompts users insert into the general chat
• Magic Prompt: Turns short or incomplete input into a well-formed prompt, with undo
• Group-based rollout: Deliver apps, skills, models, and prompts to specific departments or roles
• Portable configuration: Export an app as JSON and re-create it on another installation
• No-code authoring: Create and edit everything through the admin UI or a raw JSON editor

• Web search: Brave Search, DuckDuckGo
• Web extraction: Clean content from any webpage
• Deep research: Multi-step iterative research with synthesis
• Screenshots: Playwright and Selenium-based page capture
• File processing: Upload and analyze PDFs, text, images, audio
• Microsoft Entra: Corporate directory and people search
• Jira integration: Issue tracking and project management
• MCP: Connect further systems via the Model Context Protocol
• Self-hosted speech-to-text: Realtime transcription on your own GPU host — audio never reaches a third party
• Outlook add-in & browser extension: Centrally distributable, group-based rollout

• Multi-mode auth: Anonymous, Local, OIDC, Proxy (JWT) — mix and match
• SSO ready: Google, Microsoft, Okta, Keycloak, any OIDC provider
• Group permissions: Hierarchical group inheritance with granular access control
• Encrypted secrets: AES-256-GCM encryption for stored credentials
• CORS support: Embed iHub in other web applications

• React SPA: Built with Vite + Tailwind CSS for blazing-fast UI
• Dark/light mode: Automatic theme detection and manual switching
• Mobile-friendly: Responsive design for all screen sizes
• Multi-language: English, German, extensible to any language
• Dynamic pages: React components and Markdown content pages
• Admin panel: Full configuration management without code changes

• Zero-config startup: Auto-generates configuration on first run
• Hot reload: Config changes apply without server restart
• Multi-worker: Process clustering for production throughput
• Config migrations: Versioned, Flyway-style migration system
• Health endpoint: /api/health for load balancer probes
• Audit logging: Structured request and response logging

iHub Apps is a full-stack Node.js + React application:

• Server (/server): Express.js REST API with LLM adapters, auth middleware, and config management
• Client (/client): React/Vite SPA with Tailwind CSS, real-time streaming, and admin interface
• Configuration (contents/): JSON files for apps, models, groups, and platform settings — fully admin-editable

Request flow: Browser → Express → LLM Adapter → Provider API → Streaming SSE → Browser

The server is stateless — all configuration lives in the contents/ directory, making it easy to mount, back up, and version-control your settings separately from the application.

Contributions, issues, and feature requests are welcome! See CODE_OF_CONDUCT.md for community guidelines.

• 🐛 Report bugs: GitHub Issues
• 💡 Request features: GitHub Issues
• 📖 Full documentation: docs/README.md
• 💬 Enterprise inquiries: sales@intrafind.com

Built with ❤️ by IntraFind Software AG — Berlin · Bonn · Munich · Remote