Khoj
khoj-ai/khojReviewed by AI
Checked by RepoPicks Radar on
What is Khoj?
Khoj is a personal AI assistant you can run on your own machine. You point it at your files (PDF, Word, Markdown, org-mode, Notion, images) and it searches them and answers questions, alone or together with results from the web. You can build agents with their own instructions and tools, schedule research that arrives as a newsletter, generate images and talk to it. It works with online models (GPT, Claude, Gemini) or local ones through Ollama. Self-hosting means running several containers; there is also a hosted version at app.khoj.dev.
How easy is it to install Khoj?
A · download and open · Web
- Windows D
- macOS D
- Linux D
- Web A
The hosted version earns the best grade: the README says "You can use Khoj right away at https://app.khoj.dev — no setup required". Running it yourself on Windows, macOS or Linux is a different job: the docker-compose.yml in this package starts five services (the server, a Postgres database with pgvector, a SearxNG search engine, a Python sandbox and an optional "computer" container), and you edit environment variables by hand to add model keys and change the default passwords.
Khoj on GitHub
- Stars
- 38,100on Nov 10, 2026
- Downloads
- 186,149382 this week
- Growth
- +130stars in 7 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux, Web
- You need to install
- Nothing extra
- To build from source
- Python, Docker, Graphics card
- Disk space
- 74 MB–223 MB · from the installer
- Accounts or keys
- OpenAI, Anthropic, Google Gemini, Serper, Firecrawl, Exa or Olostep (web search and page reading), E2B (remote code sandbox)
- Cost
- Freemium · Self-hosting the AGPL-3.0 code is free; the hosted app at app.khoj.dev has free and paid plans, and the repository ships Stripe and Twilio as production extras
- License
- AGPL-3.0 (Copyleft)
- Can replace
- ChatGPT Plus, Perplexity, Notion AI, NotebookLM
- Last change on GitHub
- Aug 2, 2026
Khoj risk signals (6)
- Attentiontelemetry on by default
docker-compose.yml leaves the opt-out commented: "# Uncomment the line below to disable telemetry. Telemetry helps us prioritize feature development and understand how people are using Khoj" / "# - KHOJ_TELEMETRY_DISABLE=True"
AI review
- Attentionexample settings are wide open
docker-compose.yml ships KHOJ_DJANGO_SECRET_KEY=secret, KHOJ_ADMIN_PASSWORD=password, POSTGRES_PASSWORD=postgres and runs the server with --anonymous-mode; its own comment warns "Uncomment the necessary lines below to make your instance publicly accessible. Proceed with caution, especially if you are using anonymous mode"
AI review
- Attentionoptional setup gives access to host docker
docker-compose.yml, to enable the "computer" feature: "# uncomment line below to mount docker socket to allow khoj to use its computer" / "# - /var/run/docker.sock:/var/run/docker.sock", which lets the container control Docker on the host machine
AI review
- Infoai runs code and browses for you
The compose file includes a sandbox service (ghcr.io/khoj-ai/terrarium) "used by Khoj to run code", an optional remote E2B sandbox, a SearxNG search service and a khoj-computer container exposing VNC on port 5900
AI review
- Infolatest releases are betas without files
The three newest releases (2.0.0-beta.28, beta.27, beta.26, all March 2026) carry no platform binaries, while the last commit is from August 2026
AI review
- Infoopen core with paid cloud
pyproject.toml has a "prod" extra with stripe and twilio, and the README offers "Khoj Enterprise" as a cloud, on-premises or hybrid service
AI review
Evidence is shown in its original language.
What Radar could not check about Khoj
- What the hosted plans cost and what the free one allows
- What the telemetry sends when it is left on
- Real disk use and memory after the search and transcription models download
- Which files the release pages for Windows, macOS and Linux contain and whether they still work
- How well answers from your own documents hold up on a large collection