Your knowledge and retrieval

Your knowledge is the private set of documents you upload so the assistant can read them. A saved file is made searchable, and the assistant pulls the relevant parts into an answer, or you attach a file directly to one message. This page explains what the knowledge panel shows, the controls you set, how to read an upload's status, and what retrieval means. Knowledge and retrieval are part of the AI Console and are available on the Free, Pro Lite, and Pro plans.

Before you begin

Sign in to your account. Your knowledge is private to you.
Have a file ready in a supported format. The accepted types and the size limit are listed under Inputs.

Quickstart

Open your knowledge
Go to the AI Console and open the knowledge panel.
Upload a file
Open the upload dialog and choose a file. It moves through saving, then extracting, then becoming searchable.
Set how the assistant uses it
Pin the item to always include it, or leave it searchable so it is pulled in only when relevant.
Use it in a chat
Ask a question the file can answer, or attach the file to a single message for a one-off question.

What it shows

The knowledge area opens from the AI Console with three tabs and a per-item editor. The parts are described in the following sections.

Memory files

A file-browser layout. The left rail lists your saved items with a search box, a New text memory button, and an Upload memory file button. The right pane is an editor for the selected item. For each saved file you set a Name, a Category, a Description ("what is this file about"), and a When-to-use note ("which questions should pull this in"). The Description and the When-to-use text steer when the assistant recalls the file. An Auto-fill with AI or Regenerate with AI helper reads the file and suggests these fields for you to edit before saving.

Per-item controls are Pin (always include this item in every chat), an Active or Disabled toggle (a disabled item is ignored), Delete, and Move (share with all your subagents, or keep private to one).

File preview

A preview pane with two views. Original renders the PDF, image, spreadsheet, or markdown inline. Chunks shows how the document was split into searchable pieces and how many are indexed. A small "characters indexed" readout tells you how much of the file is searchable.

Chat memory

Automatic short summaries of your past chats, which the assistant can also search and recall. See Memory for how chat summaries refresh.

Knowledge graph

When this view is enabled for your account, it shows the entities and facts the assistant has noted, with a link back to the chat each came from.

Recall in a chat

When the assistant pulls from your library, it shows an inline "Recalled N memories" line you can expand to see each hit: the source item name and an excerpt, with a thumbs-up or thumbs-down control and an "Open in Memory Manager" jump. A separate Citations button at the bottom of a chat opens a document listing every source and tool the conversation used.

Inputs

New text memory: a free-form note up to 20,000 characters. A typical use is preferences, rules, watchlists, or standing context, for example "I prefer dividend-growth names" or "default to my Core Growth portfolio".
Upload memory file: accepted types are PDF, Word (.doc and .docx), Excel (.xls and .xlsx), PowerPoint (.ppt and .pptx), Markdown (.md and .markdown), and images (.jpg, .jpeg, .png, .webp, and .gif). The size limit is 50 MB per file. An unsupported type or an unreadable image format is rejected with guidance.
Per-file fields: Name (up to 80 characters), Category (up to 40), Description (up to 1,000), and When-to-use (up to 1,000). A name must be unique within its destination. A duplicate name is blocked with a rename prompt.
Destination: "All Memories" (shared with all your subagents, the default and recommended choice) or a specific subagent (keeps the item private to that one).
Pin: pinned items are always included. The default is off. An unpinned item is retrieved only when it is relevant.
Active toggle: the default is on. Turning it off makes the assistant ignore the item.
Attach to message: include a specific saved file for one turn. See Attachments and file uploads.
Retrieval feedback: a thumbs-up or thumbs-down on a recalled item. This is optional, and you can select again to clear it.

Outputs and how to read them

A saved item with a status: the preview's Chunks view shows indexing progress, for example "Indexing 6/12 chunks ready". Once finished, the item is fully searchable. While indexing is in progress, the assistant can still read the full extracted text.
The "Recalled" line in a reply: "Recalled 3 memories" expands to one row per hit, showing the source type (memory file, past chat, or knowledge graph), the item name, and a short excerpt of what was matched. This is how retrieval appears to you, so you can verify the assistant used the right source.
The Citations document: a per-chat list of every saved item, search, and tool the conversation drew on, with expandable detail.
Quota readouts: when your saved files approach the storage limit, or pinned items approach a size threshold, a small "File storage" and "Pinned tokens" meter appears so you know you are near a limit.

A pinned item is always available. A searchable item is retrieved only when it is relevant to your question.

How it is calculated

There is no numeric formula here, but it helps to know what retrieval means. Retrieval works by meaning, not by keyword: each saved item is split into pieces, each piece is turned into a semantic representation, and your question is matched against those pieces, so the most relevant items surface even when the wording differs. Pinned items skip that relevance step and are always included. One limitation: a very large, dense file can take a moment to finish extracting before it can be searched, so a question asked the instant after upload may not see it yet.

Data and timing

Your knowledge is entirely your own: saved notes, uploaded documents, and past-chat summaries. Nothing in this library comes from AlphaVantage, EDGAR, or Nasdaq-via-AlphaVantage. A new upload is read and indexed within seconds to a minute, longer for a large, dense PDF, and the preview shows live progress. Chat-memory summaries refresh after assistant turns once a chat has enough new activity. Saved items persist across chats and sessions until you edit or delete them.

Use cases

Save your investing preferences and rules once, so every chat answers in your style and constraints without re-explaining.
Upload reference documents (a fund prospectus, a research report, your own model notes) and have the assistant cite them in later analysis.
Keep a per-topic file, for example one company's filing notes, and let the assistant pull only the relevant one when you ask about that topic.
Use the thumbs feedback to teach the assistant which recalled sources were actually helpful.

Limitations and disclosures

Upload only material you have the right to use. Any analysis built on your documents is research, not a recommendation. Answers that read your knowledge draw down GNG Credits like any other answer.

GNG Research provides equity research and educational tools, not investment advice. Nothing on the platform is a recommendation to buy or sell any security. Do your own research and consider your circumstances before making any investment decision.