Documents & Auto-Ingest

Beyond what Pia picks up from your conversations, you can feed it your own documents. Drop files into your vault’s sources folder and Pia automatically compiles them into recallable knowledge pages the assistant can draw on in chat.

This builds on Pia’s Memory vault — worth reading that guide first.

How it works

Your vault has two layers:

• sources/ — the original documents you add. Pia reads these but never changes them.
• memory/topics/ — the knowledge pages Pia writes. For each notable subject it finds across your sources, it creates one page (a “topic”) and keeps it up to date.

When you add a document, Pia reads it, identifies the notable subjects it mentions — people, organizations, products, concepts, technologies — and writes or updates a topic page for each, combining every source that mentions that subject into one coherent page. Those pages are then searchable and citable in conversation.

An AI provider is required

Compiling documents into topic pages uses your configured AI model. Make sure you’ve set up an AI provider (or signed in to Pia Cloud) first.

Adding documents

There are two ways to add sources:

1. Drag and drop — open the Memory view and drop files onto the Vault at a glance panel on the right. You’ll see a “Drop text files here to add them to your sources” hint. Pia copies them into the sources folder and starts compiling right away.
2. Copy them into the folder — click Open memory vault in the Memory view (or the memory help dialog) to open the vault folder, then drop files into its sources subfolder. Pia notices new files and compiles them automatically.

Supported file types

Pia ingests text-based files: .txt, .md, .markdown, .csv, .json, .log, .html, .xml, and similar. Binary formats like PDF and images aren’t supported yet — those files are kept in sources/ but skipped, and shown as “Not a text file — cannot be ingested.”

Seeing what’s been ingested

In the Memory view, the Vault at a glance panel lists your Source documents. Each file shows a status:

Status	Meaning
Compiled into N topic page(s)	Ingested — its knowledge is now in your topic pages.
Not ingested yet	Queued, waiting for an AI provider, or scanned but no notable topics were found in it.
Ingesting…	Being compiled right now.
Not a text file — cannot be ingested	An unsupported format (e.g. PDF).

Re-adding the same file won’t create duplicates — Pia only re-compiles when a document’s contents actually change.

Removing a source

Delete a file from the sources folder and Pia reconciles automatically:

• Topic pages that relied only on that source are removed.
• Pages that draw on other sources too are re-synthesized from what’s left.

Note

If no AI provider is available at the time, affected pages are left in place and marked stale — Pia refreshes them the next time it compiles.

Using ingested knowledge in chat

Once a document is ingested, its topic pages behave like any other memory:

• The assistant can recall them automatically when they’re relevant to your question.
• Type @Memory to point the assistant at your memory and topic pages explicitly. See Memory → @ Commands.
• You can ask Pia directly to ingest a file — the built-in ingest tool “compiles raw documents from the vault’s sources folder into recallable memory topic pages.” It’s enabled by default; see Plugins.

Only the synthesized topic pages are added to the recall index — your raw source files aren’t embedded for search. Their contents are sent to your configured AI provider while Pia compiles them, just like any message you send in chat.

Privacy

Your source files stay in your vault folder on this device. To build topic pages, Pia sends their contents to the same AI provider you use for chat — so if that provider is cloud-based, the same privacy considerations as a normal chat apply. See the Privacy guide for details.