Building a Personal Knowledge Management Tool with Reor

We live in an era where AI technology has grown so much that it can help our everyday work. From generating simple messages and creating a long-term strategic report to automating tedious jobs, AI has become an indispensable tool everyone needs.

One AI application we can use is summarising our documents and providing additional context to our questions. We often need to pass the document manually into the model provider, such as ChatGPT, to acquire the answer. However, there is a tool that you can use locally to manage your knowledge.

This article will explore the AI tool you can use to build personal knowledge management locally.

Introducing Reor

 
Reor is an AI-powered note-taking application designed to easily take and compile notes automatically, linking any related documents. It also allows you to answer questions from your notes and perform semantic searches.

Reor works like a standard RAG system, chunking and embedding all your notes in their internal vector database using LanceDB. Then, you can perform retrieval by semantic search to answer your questions.

The Reor application can run locally, allowing you to choose from its internal embedding model. You can also select your generative LLM via Ollama or an API such as OpenAI. Additionally, you can attach your custom LLM API and embedding model.

Let’s try out the application to help you understand further.

Using Reor

 
First, you must download the Reor application for all popular OSs. Choose the one suitable for your system and proceed with the installation process. If you are successful, you will see the application shown below.

Then, you can select where you want to store your data. You can choose any folder that you feel comfortable with.

Next, you must choose which embedding model to use. You can select the recommended one or use the custom model you need. Select the default one if you are unsure which embedding model to use.

Lastly, you need to set up which generative language model to use. You have many choices: you can download models locally, use those hosted in the cloud, or use a custom LLM.

If you successfully set up the application, the greeting for using it will appear, as shown on the screen below.

Let’s try out all the features available from Reor. First, we must fill up the Reor application with all our notes files. You can also create a new folder to separate them better.

For example, I will create a new folder called research notes and fill it out with all the necessary related documents. Once the notes are ready, we will check the chat features to talk with our notes.

On the left side, you can see many tool features that support many actions, such as creating new notes or folders based on the chat. However, let’s try using the prompt to chat with the analysis we have without using the tools. You can also edit the database search filters if you need to.

With the chat features, Reor can follow the instructions by taking your notes as a knowledge base and generating responses based on the retrieved chunk.

Now, let’s see how to generate new notes based on existing ones.

In the example above, I used the prompt to generate a list of my thoughts about AI from across all of my notes. Doing that creates a new note, but only after we agree to execute the tool call.

Another feature we can try with Reor is AI-assisted writing. By highlighting the text and pressing space or clicking the wand symbol button, we can use AI to help with our note-taking.

As Reor is based on the retrieval system, you can perform a semantic search to find the notes containing terms related to your search. For example, if I use the keyword “artificial intelligence” word, it will find the notes containing the term using the highest similarity percentage.

And now you know the fundamentals of Reor. Try using it yourself to take advantage of this fantastic application’s productivity features.

Conclusion

 
Reor is an AI-powered note management system that runs locally to manage all of your notes and note-taking. By integrating with a RAG system, Reor can perform many features, such as linking related documents, performing semantic searches, and generating insightful responses. It also offers flexibility on which embedding and LLM models you want to use.

I hope this has helped!
 
 

Cornellius Yudha Wijaya is a data science assistant manager and data writer. While working full-time at Allianz Indonesia, he loves to share Python and data tips via social media and writing media. Cornellius writes on a variety of AI and machine learning topics.