Knowledge & Profile

Varen's memory is what separates it from a chat tool that forgets everything when you close the tab. This guide explains how facts are captured, where they come from, and how to use your profile to give every conversation a running start.

How Varen's memory works

Every workspace has a knowledge base — a set of structured facts that persist across sessions. When you start a new conversation in a workspace, the relevant facts are already there. You don't need to re-brief Varen on your situation, re-upload context you've already shared, or repeat facts you've already stated. The workspace accumulates this knowledge and applies it automatically.

This accumulation is deliberate and its benefit compounds over time. A tax workspace that saw you mention a property purchase in January still knows about it in June. A health workspace that noted a medication you mentioned is still aware of it six months later. The longer you use a workspace, the more precisely Varen understands the specific context of that area of your life — and the better the responses become as a result.

Fact origins

Facts come from four different sources. Understanding the difference between them helps you interpret what you see in the knowledge base and know what you can edit.

The Knowledge section lists your Stated and Insight facts. Profile and derived facts are managed elsewhere (Profile from Node Settings; the workspace intent is shown read-only in Knowledge) and are not editable from the Knowledge fact list.

Stated facts are direct assertions — you or Varen explicitly captured something you said during a conversation. They're the most reliable type of fact because they represent things you've confirmed directly. Stated facts can be edited (to update the text) and deleted.

Insight facts are inferences. Varen observed a pattern in the conversation and noted it as a likely fact, without asking you to confirm it. An insight might be something like "appears to prefer conservative investment strategies" based on how you've framed questions, or "likely self-employed" based on how you discussed income. Insights are useful context but should be treated as estimates — they're not confirmed facts, and they can be wrong. You can view them but cannot directly edit them; if an insight is incorrect, stating the correct fact explicitly will cause the incorrect inference to be superseded.

Profile facts come from your profile statement (see below). Unlike other facts, profile facts are shared across every workspace on your node — they represent the stable, persistent context that should inform all conversations regardless of which workspace you're in.

Derived facts are system-managed — the workspace name and derived intent fall into this category. They're maintained by Varen and shown read-only where relevant (for example, the workspace intent appears in the Knowledge section). They aren't edited or deleted from the Knowledge fact list.

Viewing and editing facts

Open the workspace's Knowledge section to review indexed sources, chunks, vector points, the stated and insight facts, and the workspace intent. Stated facts can be edited (pencil icon) and deleted (trash icon); insight facts are read-only. Profile is separate: open the node's settings and choose Profile. Deletion removes a record rather than crossing it out.

Stated facts support tags — short labels you add to help organise and find related facts. Tags don't affect how Varen uses the facts; they're purely for your own organisation.

Deduplication

Varen actively avoids accumulating contradictory facts. When a new fact conflicts with an existing one — you change your address, your employment situation changes, a project completes, a property is sold — the old fact is replaced rather than left alongside the new one. This is important because contradictory facts produce worse answers. A knowledge base that says both "renting" and "owns property" produces confused advice on anything related to tenancy or renovation.

If you notice an outdated fact that hasn't been automatically superseded, edit or delete it directly from the Knowledge tab. It's worth reviewing your workspace knowledge periodically — especially after significant changes in the area the workspace covers.

Profile

Your profile is a free-text statement about who you are. Open the node in the navigator, open Node Settings, and choose Profile. Unlike workspace-specific facts, profile facts are shared across every workspace on your node. They represent the stable, persistent context that should inform all conversations — regardless of which workspace you're in or what topic you're discussing.

The profile page shows the editable Profile statement, extracted Profile facts, facts Learned from conversation, User preferences, and Model-specific adjustments. These provide context when applicable; they do not guarantee that every answer uses every record.

Writing a good profile

Write prose, not a list. A few well-constructed sentences is enough. Include the things that are true across all your conversations: where you live, what you do for work, your household situation, any major ongoing circumstances that affect many areas of your life. Leave out project-specific details — those belong in the workspace knowledge base, not the profile.

As an example, a profile like this covers a great deal of useful ground in a few sentences:

I'm a 42-year-old nurse working at a public hospital in Brisbane. I own a house in Kenmore and rent out a unit in West End. I have two primary-school-age children and my mother lives with us. I run a small business providing private wound-care services on weekends, registered as a sole trader.

This level of detail means every adviser — financial, health, legal, property, government — already knows your context without you having to re-state it. A question about Centrelink is answered knowing you're employed. A tax question is answered knowing you have rental income and a side business. A health question is answered knowing you're a healthcare professional. All from a single profile statement.

What happens when you save your profile

When you click Update, Varen reads the statement and extracts a set of structured profile facts. These replace all previous profile facts — the save is not additive. If you remove a sentence from your profile and save, the corresponding fact is gone. If you add a sentence, the new fact is added.

Conversation-learned facts survive profile statement updates. Review extracted facts and remove anything inaccurate; model-specific adjustments describe corrections for a particular answer model, not the global model catalogue. There's no manual refresh or cache to clear.