Dora

Local drive, AI assisted deep search.

About

Dora is a local search tool that allows you to search files on your local drive using natural language.

Installation

Pull Repo

git clone https://github.com/space0blaster/dora.git
cd dora

With Docker

I have provided a compose.yaml file so you have everything you need (services) to run Dora. The compose.yaml file includes Ollama and ChromaDB, that's all you need for Dora search.

There is a start script called start.sh. This script is contains all the steps you need to run the compose.yaml and the electron app. The start.sh will do the following:

1. Spin up images with the compose.yaml file.
2. Pull the default models nomic-embed-text for embeddings and artifish/llama3.2-uncensored for chat search.
3. Install the Electron app dependencies.
4. Run the Electron app.

Important: If you already have Ollama and/or ChromaDB running, please spin them down first. If not, the ports will have conflicts and it will not work. If you'd rather use your existing Ollama and/or ChromaDB services, see the Without Docker section below.

To run the start.sh script:

sh start.sh

Without Docker

Assuming you already have Ollama and ChromaDB installed and running, you can simply just run the below commands to run the electron app by itself. Please note, if you already have Ollama, be sure to pull nomic-embed-text and artifish/llama3.2-uncensored as they are used. If you'd like, you can run your own choice of embedding and chat models. Configure your config/settings modal with your models. Dora defaults to the default ports for both Ollama and ChromaDB, 14434 and 8000 respectively, and will point to localhost as a base host for both.

To run the electron app:

npm install
npm start

---

Initialization

When the program runs initially, it will pop-up the Settings modal. Below are the fields:

1. Target Directory: This is the base directory which will be the starting point for the crawler/indexing function. Default is your base user directory See note no. 1.
2. Ollama Host: The host where Ollama is located. Default is localhost.
3. Ollama Port: The port for Ollama. Default is 11434 which is the default Ollama port.
4. Chroma Host: The host where ChromaDB is located. Default is 8000 which is the default ChromaDB port.
5. Embed Model: The model used specifically to create embeddings. Default is nomic-embed-text and it will be automatically pulled if you use start.sh to install and Dora.
6. Chat Model: The model used for chat search. Default is antifish/llama3.2-uncensored and it will be pulled automatically if you use start.sh to install and run Dora.

Click the Save button and you're ready to go.

Changing Target Directory

To change your Target Directory, which is where the program starts to crawl recursively, simple click on the gear icon on top right corner and enter whatever directory you want. This will reset the crawler and embed function. Use absolute paths.

---

Packaging The App

You can package the app for yourself as an application so you have an executable. Please note, if you do this, you'll still have to run the Docker images for the services for the executable to work.

To package the app:

npm run make

This will output your executable to a new sub folder in the dora directory located in dora/out and then the folder specific to your system architecture.

---

Notes

1. If you have a lot of files here, I recommend pointing it to something less dense so you can try Dora out first without the crawler running for a long time; you can change this later.

• Allow the model to warm up on the first chat request. Especially if you're running this on a relatively weak machine.
• Dora will create an application folder call .dora in your base user directory where it will store the above-mentioned configs, chat log and the indexed files.
• I am using artifish/llama3.2-uncensored because asking some models for private files freaks them out.
• You can try and use other uncensored models, I picked this one because it's relatively small and does well locally.
• You can also change your embedding model, the default nomic-embed-text does the job fine though. You can just use the embedding models that come with ChromaDB, just make sure you specify each time you embed and query.

Roadmap

• File content embedding
  • Plaintext
  • PDFs
  • Multimodal (images)

License

MIT

Contact

You can reach me on Twitter/X here.