What it does Memory Cache is a Firefox extension that saves the current webpage as a PDF into a watched folder. That folder is symlinked into privateGPT’s source_documents directory, so your local language model can ingest whatever you browse. A shell script running inotifywait detects new files and triggers privateGPT’s ingestion pipeline. You need a patched Firefox build that adds printerSettings.silentMode to the Tabs API, or you’ll be clicking print dialogs like a civilian.

The interesting bit The whole architecture is gloriously primitive: browser → PDF → filesystem symlink → shell script → local LLM. No cloud APIs, no vector database service, just your Downloads folder doing double duty as a data pipeline. The README even calls the privateGPT version “primordial,” which is either refreshing honesty or a warning, depending on your optimism.

Key highlights

• Requires compiling a patched Firefox to enable silent PDF saving (there’s a wiki page)
• Uses inotifywait to watch for new files rather than any native messaging or browser API
• Symlink-based integration with privateGPT’s folder structure
• Fork of privateGPT maintained separately by the project author
• Explicitly labeled “experimental” by Mozilla’s emerging tech incubator (Ocho)

Caveats

• Depends on a specific, outdated checkpoint of privateGPT; the README warns the setup “uses the primordial version”
• Firefox patch is mandatory and non-trivial; without it, the extension loses its “silent” quality
• No mention of PDF text extraction quality, deduplication, or how the LLM handles multi-page web PDFs

Verdict Worth a look if you’re already running privateGPT and want a friction-minimized (not friction-free) way to feed it web content. Skip it if you expect polished UX or aren’t willing to rebuild Firefox; this is a research spike, not a product.