Khoj Review

Overview

Khoj is a self-hosted AI research and writing assistant, licensed AGPL-3.0 with 37,497 GitHub stars. The project calls itself an AI second brain, and the feature list backs that up: it chats with local or online models such as Llama 3, Qwen, Gemma, GPT, Claude, Gemini, and DeepSeek, and answers questions drawn from the open web and your own documents, including images, PDFs, Markdown, org-mode files, Word documents, and Notion exports. Semantic search surfaces relevant files from that archive quickly. Access points are wide for a self-hosted tool. There is a browser interface, a desktop app, mobile access, a WhatsApp channel, and plugins for Obsidian and Emacs. Custom agents carry their own knowledge, persona, model, and tools, and automated workflows deliver personal newsletters and smart notifications by email. Image generation and voice output are included. Self-hosting is free, and a hosted option called Khoj Cloud exists, though we found no public pricing page for it in September 2026; the main site is a thin landing shell that points at the app and the docs. For content teams the practical value is document-grounded research and drafting: chat with your own archive of briefs, reports, and notes, then turn the answers into new copy. The AGPL-3.0 license means changes to a deployed instance carry source obligations, which matters before commercial use. Khoj fits teams that want AI assistance over private documents without sending those documents to a hosted model vendor.

AI Capabilities

• Chat with local and online LLMs
• Document Q&A over images, PDFs, Markdown, Word, and Notion files
• Custom agents with personas, knowledge, and tools
• Automated research newsletters and smart notifications
• Semantic document search

Key Integrations

Best for

Content and research teams that want AI help over private document collections, with custom agents and scheduled research automation running on their own infrastructure.

Not for

Teams that want a polished hosted product with published pricing, guaranteed support, or compliance certifications. Khoj Cloud exists but publishes no price list, and the self-hosted route needs technical ownership.

Review notes

Assessed from the GitHub repository, the project docs, and the hosted app rather than a full self-hosted deployment. Khoj indexes your documents, including PDFs, images, Markdown, org-mode, Word files, and Notion exports, then answers questions against that index with whichever model you connect. Local models through Ollama-style setups and hosted models from OpenAI, Anthropic, and Google are both supported, so a team can start with a hosted model and move to local inference later without changing tools.

The access surface is the part that stands out. A self-hosted assistant usually means a web page and nothing else; Khoj adds a desktop app, mobile access, WhatsApp, and plugins for Obsidian and Emacs. Custom agents take their own persona, knowledge base, model, and tools, and scheduled automations send research newsletters and smart notifications by email. That turns it from a chat window into a small content workflow engine, which is where the marketing use case lives: watch topics, gather sources, draft from your own archive.

The commercial picture is thin. Self-hosting is free under AGPL-3.0, Khoj Cloud exists as a hosted app, and no public pricing page surfaced in September 2026. The AGPL terms mean anyone deploying a modified version for external users owes source access, so legal review is worth the hour before it touches client work.

Verdict

A strong pick when the documents are the product: private research, drafting against your own archive, and automated topic watching, all on infrastructure you control. Budget setup time and model costs, and settle the license question before commercial deployment.

Similar Tools