How your agent remembers
Your agent's three memory layers — the current conversation, its own private memory, and the shared Notebook — and how each one makes answers more yours.
Most AI chat resets to zero every time you open it. Your agent doesn't. It carries what it learns about you forward — your preferences, your portfolio context, your open theses, the decisions you've made — so the hundredth conversation is better than the first.
Memory works in three layers.
Layer 1: session context
The conversation you're in right now. Everything said in the current chat session — your questions, the agent's research, files you've attached — is available to the agent for the rest of that session. Ask a follow-up and it knows exactly what you're both talking about.
Session context is the shortest-lived layer. When something from a conversation deserves to outlive it, it moves into one of the next two layers.
Layer 2: your agent's own memory
This is the agent's private, long-term memory. It has two parts:
- Working notes — the agent's day-to-day record of what it did and learned while working for you.
- Curated long-term memory — a distilled picture of you: your preferences ("short, numbers-first answers"), your portfolio context ("core position in ETH, experimenting with L2s"), and the decisions you've made along the way.
You shape this memory directly in conversation:
- Say "remember this" and it sticks. "Remember that I never invest in tokens under $50M market cap" becomes part of how the agent thinks about every future request.
- Corrections stick too. "Actually, I sold that in May" — the agent updates its picture of your holdings and stops treating the old fact as true.
Two things happen behind the scenes:
- A weekly curation pass consolidates automatically. Once a week, the agent reviews its working notes and folds what matters into long-term memory, so important facts don't get buried under day-to-day noise.
- It can search its own past semantically. Ask "what did we conclude about that gaming token a while back?" and the agent can find the earlier work by meaning, not just exact words.
Layer 3: the shared Notebook
The Notebook is durable memory that both of you read and write — saved ideas, analyses, frameworks, and verdicts, each with a stable NB-number you can reference in chat. It's where a conversation's conclusions become a record you can return to.
Memory vs. Notebook
The agent's memory and the Notebook solve different problems:
| Your agent's memory | The Notebook | |
|---|---|---|
| Who sees it | Private to the agent | Both of you — read and edit |
| How it's maintained | Automatic (plus "remember this") | Curated together; the agent asks before saving |
| Where it's visible | Surfaces in the agent's answers | At blockmind.app/elite/notebook and a dashboard widget |
A rough rule: memory is how the agent understands you; the Notebook is what you've decided together.
Why this matters
- Answers get more yours over time. The agent stops giving generic research and starts answering the way you'd want a long-time analyst to — in your terms, aware of your positions and risk tolerance.
- Briefs and research reflect your open theses. Your Morning Brief and research sessions pick up where your thinking left off, instead of starting from a blank page.
- Nothing is lost between sessions. Close the tab, come back next week — the context is still there.
Everything in these layers is private to your workspace. Your conversations, memory, and Notebook are never shared with other users.
Related
Custom capabilities
Teach your agent your own repeatable workflows in conversation. It drafts the capability, saves it only after you approve, and runs it whenever you ask.
The Notebook
The shared Notebook holds durable notes with stable NB-numbers — ideas, analyses, frameworks, and verdicts you and your agent build together.