Khoj

Local or hosted model choice Chat with on-device models such as Llama, Qwen, Gemma, and Mistral through an Ollama-compatible endpoint, or switch to GPT, Claude, and Gemini with an API key. The self-host guide shows exactly where each key goes, so a household can start free on local models and add commercial ones later.[1][2]

Semantic search over personal docs Find relevant notes fast across PDFs, Markdown, org-mode, Word, Notion, and images instead of relying on keyword search. Because retrieval understands meaning, a half-remembered idea surfaces even when the exact words do not match.[1]

Custom agents with roles Create agents with their own knowledge, persona, chat model, and tools to take on research, tutoring, or support roles. A server host can set up agents once and make them available to everyone on that server.[1][2]

Scheduled automations Run a query automatically at a set time and frequency, with results delivered as email newsletters and smart notifications. Repetitive routines happen without retyping the same prompt.[1][2]

Online search with references Khoj researches the internet to ground answers when its own knowledge is not enough, and always shows the references it used. Readers can verify a claim against the cited pages instead of taking the answer on trust.[1][2]

Python code execution Khoj generates and runs simple Python code for data analysis, plots, and reports that pure language models handle poorly. Quantitative questions get computed answers rather than guessed ones.[1]

Voice and image generation Talk out loud, play messages back, and generate images from text prompts without leaving the assistant. A desktop shortcut opens a quick mini-chat for short questions that should not break flow.[1][2]

Clients everywhere Reach the same assistant from the browser, desktop, phone, and WhatsApp, or inside Obsidian and Emacs, and share any conversation with a public link. Shared links freeze the conversation at the moment of sharing, so later messages stay private.[1][2]

People with large Obsidian or Notion libraries get direct answers from their own writing instead of re-reading folders. The Emacs and Obsidian clients matter here because they answer inside the tool where the notes already live.[1][2]

Anyone with a home server or spare computer keeps documents on their own network, including fully offline use. Pairing that with local models through Ollama removes the API-key trail as well.[1][2]

People who enjoy setting up agents and scheduled digests get recurring research without repeating the same prompts. Email delivery closes the loop for readers who do not want another dashboard to check.[1][2]

Khoj is personal software; teams wanting isolated users, roles, and admin controls fit AnythingLLM better, since its workspace and role model is built for sharing.[1][2]

Single sign-on, audit logs, and data-residency paperwork were not documented on the inspected pages, so regulated buyers cannot tick those boxes from this review.[1][2]

Self-hosting is free apart from infrastructure and model keys, but per-seat cloud prices were not published on the inspected pages, so budget owners should confirm numbers in the live app first.[1][2]