Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Khoj

Self-hostable AI second brain: chat over your notes and documents with local or hosted models, now self-host only after its cloud closed.

Reviewed by Sheela · Updated

View on GitHub Open source

Our rating

3.2/5

Khoj was one of the few second-brain apps that put Obsidian and Emacs users first, and with its cloud gone and its makers elsewhere it is now a self-host project to adopt with eyes open.

What is Khoj?

Khoj Cloud, the hosted version at app.khoj.dev, shut down on 15 April 2026. The founders say they are now focusing on Open Paper, a literature-review assistant, and Pipali, a desktop AI coworker. Khoj itself stays open source and self-hostable, but development has slowed. The last commit on the main repo was on 2 August 2026, the latest GitHub release is a 2.0.0 beta from 26 March 2026, and the latest stable PyPI release, 1.42.10, dates from 15 July 2025.

Khoj is a personal AI app. It indexes your notes and files (PDF, Markdown, org-mode, Word, Notion, images), answers questions from them and from the web with any local or hosted model, runs custom agents and scheduled research, and is reached from a browser, Obsidian, Emacs, the desktop or WhatsApp. It is AGPL-3.0 Python, with about 37,900 GitHub stars. It installs with Docker Compose or pip install 'khoj[local]' and uses a Postgres database.

Is Khoj worth it?

The facts that matter most are about the project, not the features. The team closed Khoj Cloud in April 2026, saying a subscription, cloud-first service with six clients and hand-built data integrations was hard to scale. They have moved to other products. The repository still gets fixes: an August 2026 commit stopped sending client IPs in telemetry so behaviour matched the privacy docs. But the stable PyPI release is more than a year old and the 2.0 line has stayed in beta since March.

What remains is still distinctive. Few self-hosted assistants reach into Emacs and Obsidian as first-class clients, or combine semantic search over org-mode and Markdown notes with web research, custom agents and scheduled newsletters. Offline chat runs on open-weight models through any OpenAI-compatible local server. The docs give a minimum of 8 GB RAM, recommend 16 GB VRAM, and need 5 GB of disk.

The security record is modest in size but recent. Six advisories have been published, four medium and two low, including a June 2026 unauthenticated path traversal that allowed reading files from the server. AGPL-3.0 also matters for anyone thinking of offering Khoj to others over a network: modifications must be shared with those users.

Install friction is creeping in as the code ages. The PyPI package requires Python below 3.13, and an open October 2026 issue reports pip installs failing on Debian 13. Docker Compose is the safer route. For new deployments, AnythingLLM and Open WebUI are more actively maintained. Khoj makes sense mainly for people already invested in its Emacs or Obsidian workflow.

Where it earns its place

  • A researcher with ten years of org-mode notes in Emacs

    Ask questions across old notes without leaving Emacs

    Self-host the server with Docker Compose, install the khoj.el package and index the notes directory. Free, and private when paired with a local model; the trade-off is maintaining a project whose stable release is over a year old.

  • A former Khoj Cloud subscriber

    Keep using Khoj after the hosted service closed in April 2026

    Move to the self-hosted server with Docker Compose and re-index the documents, since the cloud is gone. A small VM or home server suffices with a hosted model; offline chat wants 16 GB VRAM by the docs' recommendation.

  • An IT lead at a 15-person firm comparing self-hosted assistants

    Choose a private document chat to support for three years

    Weigh Khoj's clients against a slowing release cadence, six published advisories and AGPL-3.0 obligations. For a long-lived deployment, AnythingLLM or Open WebUI carry less maintenance risk.

Before you commit

Setup
Download the docker-compose.yml into ~/.khoj and run docker-compose up, or pip install 'khoj[local]' (Python 3.10 to 3.12) with build flags for CUDA, ROCm, Vulkan or Metal. Restart after the first run, per the docs.
Integrations
Clients for browser, Obsidian, Emacs, desktop and WhatsApp; Notion and file indexing; local OpenAI-compatible model servers and hosted models from OpenAI, Anthropic, Google and others.
Data and privacy
Self-hosted, so notes stay on your server unless you use a hosted model. Telemetry exists; an August 2026 fix stopped it sending client IPs. Six GitHub advisories, including a June 2026 unauthenticated path traversal. AGPL-3.0.
Cost at scale
Free software; no hosted option since April 2026. Offline chat needs at least 8 GB RAM, with 16 GB VRAM recommended; with a hosted model a small VM is enough.
Limits
Stable PyPI release 1.42.10 is from July 2025; 2.0 is still beta. PyPI requires Python below 3.13, and a Debian 13 install failure is open. Maintainer attention has moved to Open Paper and Pipali.

What’s good

  • First-class Obsidian, Emacs, desktop and WhatsApp clients
  • Works with local open-weight or hosted models
  • Semantic search over notes plus web research and agents
  • Self-hosting keeps notes on your own server

What’s not

  • Hosted cloud shut down 15 April 2026
  • Stable release dates from July 2025; 2.0 still beta
  • Maintainers now focus on other products
  • AGPL-3.0; PyPI build needs Python below 3.13

Best for

Existing Khoj users and Emacs or Obsidian note-takers willing to self-host a slowing project for a private, model-agnostic second brain.

Khoj alternatives

Khoj: common questions

Is Khoj worth it?
Khoj was one of the few second-brain apps that put Obsidian and Emacs users first, and with its cloud gone and its makers elsewhere it is now a self-host project to adopt with eyes open. We rate it 3.2 out of 5.
How much does Khoj cost?
Free and open source (AGPL-3.0), self-hosted only since Khoj Cloud closed on 15 April 2026.
What are the downsides of Khoj?
Hosted cloud shut down 15 April 2026; stable release dates from July 2025; 2.0 still beta; maintainers now focus on other products; aGPL-3.0; PyPI build needs Python below 3.13.
Who is Khoj best for?
Existing Khoj users and Emacs or Obsidian note-takers willing to self-host a slowing project for a private, model-agnostic second brain.
How much does Khoj cost for a team or in production?
Free software; no hosted option since April 2026. Offline chat needs at least 8 GB RAM, with 16 GB VRAM recommended; with a hosted model a small VM is enough.
Is Khoj safe for company data?
Self-hosted, so notes stay on your server unless you use a hosted model. Telemetry exists; an August 2026 fix stopped it sending client IPs. Six GitHub advisories, including a June 2026 unauthenticated path traversal. AGPL-3.0.
What are the best Khoj alternatives?
The closest alternatives we have reviewed are AnythingLLM, Open WebUI, LibreChat and claude-obsidian.

Answers checked Oct 31, 2026.