Khoj AI - Self-Hostable AI Research App

Source: https://www.makeuseof.com/started-using-self-hostable-app-for-research-should-have-sooner/ Summarized: 2026-02-23

TL;DR

Khoj AI is a middle ground between ChatGPT (too minimal) and NotebookLM (too heavy). Self-hostable, supports custom agents, automations, and your own models via Ollama. Think of it as "NotebookLM + Claude had a baby."

What is Khoj AI?

A research assistant that combines web search, document analysis, and LLM chat. Two ways to use:

Cloud: Free tier with Gemini Flash 3 and basic models
Self-hosted: Docker + bring your own model (Ollama supported)

Key Features

1. Built-in Agents

Pre-configured personas:

Khoj (default)
Technical Lead
Teacher
Legal Expert

Switch agents per conversation for role-specific responses.

2. Slash Commands

Command	Function
`/notes`	Pull info only from your uploaded documents
`/code`	Launch built-in Python interpreter (can generate graphs via Matplotlib)
`/web`	Web search integration

3. Custom Agents

Create your own:

Add files to knowledge base
Choose model
Set input/output modes
Done

4. Automations

Schedule recurring tasks:

Daily stock market summaries at 9 AM
RSS feed fetching at set times
Results delivered to email automatically

No code required.

Self-Hosting Setup

Requirements: Docker + decent hardware (local LLMs need beefy machines)

mkdir ~/.khoj && cd ~/.khoj
wget https://raw.githubusercontent.com/khoj-ai/khoj/main/docker-compose.yml
nano docker-compose.yml  # Set admin email/password, add API keys
docker-compose up

Access: http://localhost:3600

Model options:

Use third-party providers (OpenAI, Anthropic, Gemini) with API keys
Use local models via Ollama

Why Choose Khoj Over NotebookLM?

Khoj	NotebookLM
Self-hostable	Cloud only
Custom agents	Fixed structure
Automations	Manual queries
Bring your own model	Google models only
Middle ground complexity	Heavy, structured

Use Cases

Students: Research, understanding topics (not copy-pasting assignments)
Work: Document analysis, research workflows
Personal projects: Custom agents for specific domains

Caveats

LLMs can hallucinate — always verify important info (legal, medical)
Local models need strong hardware
Accuracy depends on model choice

Bottom Line

Khoj fills the gap between minimal chat interfaces and heavy research tools. Self-hosting gives you full stack ownership—own, don't rent.

2.6 KiB Raw Blame History