Home

Memory Cache is the working name for a set of projects and tools that enable you to turn your local desktop computer into an AI agent that can provide insights and enable the ability to query information from a "cache" of documents provided to a locally-running instance of privateGPT. Over time, the hope is that this project will evolve into a standalone executable application that makes local AI use and experimentation more accessible.

Components of Memory Cache:

• MemoryCacheExt (Firefox Extension) - captures content from a browser extension that saves it into a symlinked folder to privateGPT
• Custom Firefox capabilities (Firefox source code) - ext-tabs.js modification adds the capability for the tabs.saveAsPDF API to run in silentMode to reduce friction of saving content from the browser.
• privateGPT - an open source project that includes an ingest.py script that allows for search-augmented retrieval capabilities by taking in new documents and files

Hardware

I'm running Memory Cache on my primary desktop computer. See [https://github.com/misslivirose/Memory-Cache/wiki/Supported-Memory-Cache-Builds] for more details on hardware.

Querying privateGPT on 'Ori' takes slightly over one minute, depending on the complexity of the query. The output includes a generated attempt at answering the question, and four references to related documents.

Paths explored, but not taken:

• history-plus patch stack for Firefox - dumps an extended history into a local .sqlite database to be used as a source of fine-tuning for local AI model