At a Glance:
Khoj is an open-source, self-hostable personal AI app that scales from on-device use to cloud enterprise deployments, supporting chat with local or online LLMs, semantic document search, custom AI agents, and multi-platform access including browser, Obsidian, desktop, and WhatsApp.
Overview:
Khoj is a personal AI application designed to extend user capabilities by acting as a second brain that scales from an on-device assistant to a cloud-scale enterprise AI. It enables users to chat with any local or online large language model, including llama3, gpt, claude, and gemini, and to get answers from both the internet and personal documents such as PDFs, markdown, Notion pages, and Word files. The application supports creating custom AI agents with distinct personas, knowledge bases, and tools, and can automate repetitive research through personal newsletters and notifications. Khoj is accessible across a wide range of interfaces, including web browsers, Obsidian, Emacs, desktop applications, phones, and WhatsApp, and offers capabilities such as image generation, text-to-speech, and advanced semantic search for locating relevant documents.
Key Decision Points:
Deployment flexibility: Can be self-hosted privately on a personal computer or used immediately via a managed cloud app.
LLM model choice: Supports connecting to both local models (llama3, qwen, gemma, mistral) and online LLM services (gpt, claude, gemini, deepseek).
Document ingestion scope: Reads a wide variety of document types including PDFs, markdown, org-mode, Word, and Notion files, making it suitable for users with diverse knowledge bases.
Interface accessibility: Available through multiple clients, including a web browser, Obsidian, Emacs, desktop app, phone, and WhatsApp, providing flexible access points based on user workflow.
Automation and agent capabilities: Allows users to create custom AI agents and set up automated personal newsletters and notifications, moving beyond simple chat into ongoing research automation.
Core Features:
Multi-model chat: Chat with any local or online LLM, including llama3, qwen, gemma, mistral, gpt, claude, gemini, and deepseek.
Document-grounded answers: Get answers drawn from the internet and personal documents, including images, PDF, markdown, org-mode, Word, and Notion files.
Custom AI agents: Create agents with custom knowledge, persona, chat model, and tools to take on specific roles.
Automated research and notifications: Automate repetitive research and receive personal newsletters and smart notifications delivered to an inbox.
Advanced semantic search: Quickly find relevant documents using advanced semantic search.
Multi-platform access: Access Khoj from a browser, Obsidian, Emacs, desktop, phone, or WhatsApp, with support for image generation and voice output.
Use Cases:
Developers and researchers can use Khoj to chat with local LLMs, query personal document collections, and automate repetitive research tasks across PDFs, markdown notes, and org-mode files.
Knowledge workers can create custom AI agents with specific personas and knowledge to handle specialized information retrieval or newsletter generation from their existing document bases.
Self-hosters can run Khoj entirely on their own device, connecting to locally running models for private, on-device AI assistance without relying on cloud services.
Users seeking always-available AI access can interact with Khoj through multiple interfaces including WhatsApp, Obsidian, and Emacs, fitting the AI into existing workflows rather than requiring a separate application.
Open-Source Alternative Value:
Khoj provides an open-source, always self-hostable personal AI application that supports both local and online LLMs, giving users the option to run the system privately on their own computer or use a managed cloud instance. Unlike proprietary AI assistants tied to specific model providers, Khoj allows switching between local models (llama3, qwen, gemma, mistral) and commercial APIs (gpt, claude, gemini, deepseek) based on user preference. The ability to create custom agents with distinct knowledge bases and tools, combined with semantic search across diverse document formats and multi-platform access through Obsidian, Emacs, and WhatsApp, offers a transparent and extensible AI second brain that can be adapted to individual workflows without dependency on a single vendor.
