talos

Talos is meant to be a powerful interface to easily create automated workflows that uses large language models (LLMs) to parse, extract, summarize, etc. different types of content and push it to other APIs/databases/etc.

Think Zapier but with language model magic 🪄.

Click to play video. Note: the summarization step is sped up (~1 minute).

Example Workflows

Read this Wikipedia page & extract some data from it.

View workflow

1. Fetch this wikipedia page.
2. Extract who the page is about, give a summary, and extract any topics discussed.
3. Put into a template.

Read a Yelp review & tell me about it.

1. Read this Yelp review
2. Extract the sentiment and give me a list of complaints and/or praises
3. Put into a report template.
4. Email it to me.

Summarize this book & generate a book report.

1. Read through this PDF file.
2. Create a bullet point summary of the entire book.
3. Generate key takeaways from the book 3a. For each takeaway, elaborate
4. Combine into a template.

Running Locally

To start running talos locally, install the dependencies

# Copy the environment variables
> cp .env.template .env.local
# Install dependencies
> npm install
# Start the front-end
> npm run start

The UI will be available at http://localhost:3000/playground.

Now we need to start the backend.

Using w/ memex

memex is a self-hosted LLM backend & memory store that exposes basic LLM functionality as a RESTful API.

You will need to either download a local LLM model to use or add your OpenAI to the .env.template file to get started.

> git clone https://github.com/spyglass-search/memex
> cd memex
> docker-compose up