Mozilla has released an experimental MemoryCache add-on offers a conversational machine learning AI bot. It takes into account the content accessed by the user in the web browser. Also, unlike other AI chat tools, MemoryCache enables personalized communication with the user and utilizes crucial user-specific data to generate responses to queries.
The project's code is available under the MPL license. Currently, installation in Firefox is only supported browser by the bot. It can only be activated manually on the "about:debugging" page, and the browser must be compiled with a special patch.
All MemoryCache components operate on the user's system and do not interact with external resources. MemoryCache utilizes the GPT4All-J model, serving as a knowledge base, which is trained on collections of English-language content Pile (825 GB of data), ShareGPT (13 MB), and Dolly (660 MB), offered in versions with 3, 7, and 13 billion parameters. The privateGPT toolkit is employed to work with the model, an optimized solution for employing large language models on local computers.
The Memory Cache add-on stores viewed pages in the "~/Downloads/MemoryCache" directory, where a script is periodically executed to retrieve and transmit them to the privateGPT handler (ingest.py) to supplement the existing language model.
Consequently, the model adapts to the user's interests and can be used to discuss the viewed content in a text chat with a bot.
Users are able to ask the bot questions about saved pages in natural language, such as to clarify facts, with the bot generating responses using the local model rather than external services.
Support us
Winaero greatly relies on your support. You can help the site keep bringing you interesting and useful content and software by using these options: