While NotebookLM is an amazing research tool, and there are some great prompts that can make it even more powerful, I’ve always felt like it was a bit too much for my needs, even if that sounds like an unpopular opinion. I wanted something that sits in the middle, not as minimal as the ChatGPT website, but not as heavy or structured as NotebookLM.
That’s when I started exploring Khoj, (which translates to “discovery” in English). After spending some time with it, I genuinely feel like I’ve found that perfect middle ground I was looking for.
One of NotebookLM’s most useful features is now available on Android
Video Overviews make it easier than ever to absorb information
NotebookLM is brilliant, but I like this more
Firstly, there are two ways to use Khoj AI. You can either use the website, where the free tier has access to basic models like Gemini Flash 3, or you can self-host and bring your own model. To get started, I would recommend that you try out the website first. There is no point in setting everything up locally if you don’t end up liking it.
When you open the website, it looks very similar to Claude, so you’ll feel right at home. There’s a chat box along with various features for writing code, analyzing documents, and more. Basically, it’s ready to handle most “research-heavy” tasks right away.
Above the chat box, you’ll also see multiple agents. The default is Khoj, but you can switch over to something like technical lead, a teacher, or even a legal expert. Just select which one you want and it’s ready to go!
Remember that LLMs can hallucinate, no matter how capable the model is. Always double-check important information, especially when it comes to areas like legal or medical advice.
One of the really cool things Khoj offers is slash commands. Just type “/” and you’ll see a list of commands that you can add to your prompt.
For example, typing “/notes” forces the LLM to pull information only from your own documents. Using “/code” launches the built-in Python interpreter, which can even generate graphs and other figures directly inside the app using Matplotlib. The best way to describe this app would be if NotebookLM and Claude had a baby.
You can even build your own agents and automations
Do even more than research
I mentioned earlier that Khoj AI includes built-in agents that can take on different roles, like acting as a coach or a legal expert. But what makes it even more powerful is that you can create your own agents tailored to your specific workflows.
In the Agents tab, select Create Agent, and you’ll be prompted to enter a few basic details. After that, you can add files to its knowledge base, choose the model, set the input and output modes, and you’re all set.
But obviously, as I mentioned earlier, the accuracy of your agent is largely going to depend on the model you chose. If something goes wrong, it’s more the model’s fault than Khoj AI’s.
Just like agents, you can also create your own automations that run tasks automatically at a specified time (duh.) For example, you could receive a full summary of the stock market every day at 9:00 AM, or set up an agentic-esque RSS feed that fetches the latest stories at a scheduled time.
There are plenty of pre-built automations available, but you can also create your own. Just go to the Automations tab and select Create Automation. From there, choose the days and times it should run, and specify what information you want the LLM to generate.
That’s it! You’ll receive the information at the same time every day straight to your email, without writing a single line of code.
You own the entire stack when you self-host Khoj AI
Own, don’t rent
I get it if you don’t want to rely on the cloud, and the good thing is you can self-host Khoj via Docker. It can even call your LLMs via Ollama, so that’s a huge hassle taken care of as well. To get started with running it on your own hardware, ensure you have Docker installed on your system. Once that is done, simply run the following command in your terminal:
mkdir ~/.khoj && cd ~/.khoj
wget https://raw.githubusercontent.com/khoj-ai/khoj/main/docker-compose.yml
Now, you’ll need to configure some things in the docker-compose.yml file. You can do so by entering this command:
nano docker-compose.yml
Open the file and look for KHOJ_ADMIN_EMAIL and KHOJ_ADMIN_PASSWORD. Replace them with the email and password you want to use for your admin account.
If you plan to use a third-party LLM provider, you can also add your API keys for OpenAI, Gemini, or Anthropic in their respective fields in the same file. After that, you’re pretty much done! Just ensure you’re still in the same directory in your terminal and enter the following command:
docker-compose up
By default, Khoj will open at http://localhost:3600, so enter that URL in your browser, and you’re ready to go! Just keep in mind if you’re using your own models, you might need a really beefy machine for the best performance. Running LLMs locally isn’t exactly the lightest thing.
I hooked Obsidian to a local LLM and it beats NotebookLM at its own game
My notes now talk back and it’s terrifyingly useful.
Bye bye, NotebookLM (for now)
As a student, it has been a really useful tool for me, mainly for research and understanding topics better, and obviously not for copying and pasting assignments. It works best when you use it to support your thinking rather than replace it.
You also do not have to be a student to benefit from it. You can use it for work, personal projects, or pretty much anything else you are working on. And if Khoj is not enough on its own, you can always combine it with other great web research tools to make your workflow even more powerful.
