Skip to main content
Once you have connected an AI agent to a brain, the easiest way to add anything is simply to tell it. You do not open a form or upload a file: you say what you want remembered, and the agent files it for you. Behind the scenes, the agent follows the ingest skill, the standing instruction for reading something and filing it in the right place with a citation. You never have to think about that; you just ask.

​
What to say

Plain requests work. For example:
  • “Save this to my brain: we’re switching the Acme account to annual billing.”
  • “Remember that Priya owns the onboarding project.”
  • “File these notes as a page about our refund policy.”
  • “Ingest this document and file it.” (paste the text, or point it at something in the conversation)
  • “Save a summary of what we just worked out.”
  • “What tasks have been captured this week? Keep the Perlon one and drop the rest.” (your task page, run by conversation)
The agent reads what you gave it, writes a tidy page in the right part of the brain, links it to related pages, and records where it came from, exactly as if you had filed it by hand. What your agent brings in, your agent files, there and then, with its own tokens. Cortex’s own filer never touches it. How ingestion is paid for has the detail.

​
Giving your agent a file

Hand a file to an agent that has a shell, Claude Code say, and ask it to file it. This is what happens:
  1. The agent uploads the file first. If the command line is installed, the agent uses it, and the file streams from disk into the brain’s inbox. If not, the agent asks Cortex for a one-time upload address and sends the file from its shell. Either way the file’s bytes never pass through the conversation. Only when the agent has no shell at all, and the file is small, does it send the file inline.
  2. Cortex reads the file and hands the text back. The upload answers with the new inbox item, its citable address, and the document’s text, extracted on the server from PDF, Word, PowerPoint or plain text. If the text cannot be extracted (a scan, an image), the answer says so and the agent reads the file with its own tools. If the file is a new version of a document already filed, the answer says which pages the earlier version fed and what changed.
  3. The agent writes the pages and archives the item. It writes the pages that cite the source, a few at a time, and archives the inbox item with the list of pages it wrote, so the source and the pages point at each other.
If the command line is not installed but Node is, the agent may offer once to install it for you (npm install -g @oncortex/cli). It never installs anything unasked; the upload address route needs nothing installed. When the document lives somewhere, say on SharePoint or Google Drive, the agent fetches it through your own connected tools (a private link does not open in a browser the agent drives) and records where it lives alongside the upload. The source record then shows the copy that was read and a link to the live document, and a later version of the same document is recognised as a revision. Nothing inside a document can tell the agent where to file it; only the agent, and you, decide that.

​
The first time an agent writes into a brain

Before its first write into a brain, your agent reads the brain’s instructions page: what the brain is for and how it is organised. If that page is thin, because nothing has been written, it is still the untouched starter text, it is under sixty words, or it says nothing about folders that already hold pages, the agent stops and asks you what the brain is for, or shows you a short draft to approve or edit. It saves the instructions only after you reply, and only then files. Every later filing, by the agent or by Cortex, reads that page, so a minute spent here pays back on every item.

​
Why this is the easy way

There is nothing to learn. If you can talk to your agent, you can add to your brain. It is quicker than uploading, it works for things that only exist in your conversation (a decision you just reached, a fact you mentioned), and the result is filed properly rather than dumped in raw.

​
The other ways still exist

Asking your agent is the easiest path, not the only one:
  • To add a document or file directly, without an agent, upload it.
  • To bring in meetings, push transcripts in the transcript format.
  • To bring in mail, forward a thread to your brain’s address, or ask an agent that can read your mail to file one. See your email.
  • To capture your agent conversations without asking each time, set up automatic capture.
Haven’t connected an agent yet? See Connecting your AI first; it takes a couple of minutes.

​
Try it

  1. With an agent connected, say “Save a note to my brain that I prefer morning meetings.”
  2. Open the brain and find the new page, with its source noted at the bottom.
  3. Ask the agent “what do you know about my meeting preferences?” and watch it recall what you just saved.