fossil-mastodon 0.3.0

A mastodon reader client that uses embeddings to present a consolidated view of my mastodon timeline

Fossil, a Mastodon Client for Reading

A mastodon client optimized for reading, with a configurable and hackable timeline algorithm powered by Simon Wilison's llm tool. Try making your own algorithm!

Sneak peek:

Installing & Running

From PyPi

I highly suggest not installing any Python app directly into your global Python. Create a virtual environment:

python -m venv fossil

And then activate it (see here)

source fossil/bin/activate

Alternatively, use pipx:

pip install pipx
pipx install fossil-mastodon

From Source

Clone this repo:

git clone https://github.com/tkellogg/fossil.git

And then cd fossil to get into the correct directory.

Configure the .env file

Before that, you'll need a .env file with these keys:

ACCESS_TOKEN=

Alternatively, you can set them as environment variables. All available keys are here:

Variable	Required?	Value
OPENAI_API_BASE	no	eg. https://api.openai.com/v1
MASTO_BASE	no?	eg. https://hackyderm.io
ACCESS_TOKEN	yes	In your mastodon UI, create a new "app" and copy the access token here

Connecting to Mastodon

To get MASTO_BASE and ACCESS_TOKEN:

1. Go to Mastodon web UI
2. Preferences -> Development
3. Click "New Application"
4. Set the name
5. Set "Redirect URI" to urn:ietf:wg:oauth:2.0:oob
6. Set scopes to all read and write (contribution idea: figure out what's strictly necessary and send a pull request to update this)
7. Click Submit
8. Copy your access token into ACCESS_TOKEN in the .env file.
9. Set MAST_BASE. You should be able to copy the URL from your browser and then remove the entire path (everything after /, inclusive).

Usage

1. Ensure the settings are correct
2. "Load More" to populate the database with toots
3. "Re-Train Algorithm" to categorize and label those toots.

Configure Models

Models can be configured and/or added via llm.

OpenAI

Here's how to set your OpenAI API key, which gives you access to OpenAI models:

$ llm keys set openai
Enter key: ...

Local (Experimental)

You will need to install an embedding model and a large language model. The instructions here use the llm-sentence-transformers and llm-gpt4all plugins to do so.

$ llm install llm-sentence-transformers # An Embedding Model Plugin
$ llm install llm-gpt4all # A Large Language Model Plugin
$ llm sentence-transformers register all-mpnet-base-v2 --alias mpnet # Download/Register one of the Embedding Models

Notes

• A full list of possible embedding models is composed of the default list and these models from huggingface.
• The llm-gpt4all README gives a list of models and their requirements
• The first time you use a model, llm will need to download it. This will add to the overall time it takes to process
• The "Re-Train Algorithm" step will take a long time depending on your hardware; a progress bar is shown in the console window
• The quality of the categorization and labels are not guaranteed

Run the server

If you installed from PyPi:

uvicorn --host 0.0.0.0 --port 8888 fossil_mastodon.server:app

If you installed from source:

poetry run uvicorn --host 0.0.0.0 --port 8888 --reload fossil_mastodon.server:app

If you're working on CSS or HTML files, you should include them:

poetry run uvicorn --host 0.0.0.0 --port 8888 --reload --reload-include '*.html' --reload-include '*.css' fossil_mastodon.server:app

(Note the --reload makes it much easier to develop, but is generally unneccessary if you're not developing)