txtask is a CLI application that allows you to chat with your local text files using Ollama.
It works by indexing the text files in your ./data folder and creates an embedding for each chunk of the files. All the embeddings are stored in memory.
Upon receiving an input (the question), txtask will calculate the similarity between the embedding of your question and the stored embeddings to find the matching chunks. Then, use these chunks as context to answer the question.
You will need to have Ollama installed locally, and run the Mistral 7B model:
$ ollama run mistral
You can use any model you want, but remember to set the correct model in th e OllamaClient's request payload in the code.
Clone the repository to your machine, and run with:
$ cargo run
txtask will start indexing file markdown files in ./data folder.
After that, you can start asking questions.
The embeddings are cached in the database.json file, and it will be loaded after the first run. So, if you add more text files to the ./data folder, delete this JSON file so txtask can index them.