Super Brain: shared memory for your AI team, explained

Super Brain is the shared memory of your KutrimBots AI team. Business and project knowledge, such as clients, prices, processes, FAQs and decisions, is saved once and every agent can search it. What an agent remembers on its own stays private to that agent unless you allow another one to read it. Everything is stored as a knowledge graph of Markdown notes, people and things, with no size limit, and nothing is pasted into every prompt.

This guide explains why AI agents need shared memory, how Super Brain works, and how to fill it well.

Why do AI agents forget things?

Most AI assistants remember only the current conversation, plus a few saved facts. That causes three everyday problems:

• You repeat yourself. You explain your pricing to the sales agent, then again to the support agent.
• Agents contradict each other. One quotes last year's price because it never heard about the change.
• Context windows are not memory. Pasting everything into every prompt is slow and expensive, and stops working as your business grows.

Researchers point at the same gap: a 2025 framework for measuring general intelligence in AI scored today's models at 0% on "long-term memory storage", the ability to keep learning new information. Products have to build memory around the model.

How does Super Brain work?

Team memory and personal memory

• Team memory is what agents and you saved for everyone. Every agent reads it.
• Personal memory is what an agent remembered on its own, or saved as private. Only that agent reads it.
• Asking for access: when an agent needs another agent's personal memory, it sees only that matches exist ("3 more matches are in Penny's personal memory"), never the text. It asks you, and you decide: just this once, for 24 hours, always, or no.

This keeps a customer's personal details with the agent that handled them, while shared facts like prices and processes are available to the whole team.

A knowledge graph, not a pile of documents

Each note is about things: people, companies, projects, products and topics. Things are connected by relations, such as "Priya — works at → Acme" or "Acme — client of → our studio". That lets an agent answer questions a keyword search can't:

• "Everything we know about Acme": current facts, past facts and every note connected to it.
• "How is Priya connected to the Mural project?": the path through the graph.
• "What did we decide about refunds?": the decision note, with its date.

When a fact changes, such as a price update or someone moving companies, the old fact is kept as history and the new one becomes current.

Search that forgives typos

Agents find things by:

• words;
• close spellings (typos and Hinglish are fine);
• meaning (semantic search: "cost" finds "pricing");
• the graph itself.

Results say why they matched, and agents say which note an answer came from.

Nothing pasted into every prompt

Super Brain isn't stuffed into each request. Agents look things up when they need them. That keeps answers fast and cheap, and it's why there's no size limit.

What should I put in Super Brain?

Start with the knowledge you repeat most. A good first week:

Write one topic per note with a clear title, like "Refund process" or "Acme — pricing formula". Link notes with [[Note title]], add #tags, and update a note instead of creating a near-copy.

Never put passwords, API keys or card details in Super Brain. They belong in the encrypted Security Vault, where agents use them without seeing them.

How do I add knowledge?

• Say "remember". "Remember: our delivery is free above ₹999 in Pune." The agent saves it to team memory.
• Let agents save as they work. When an agent learns a lasting fact or decision, it saves it.
• Paste or upload. Give an agent your price list, FAQ document or process doc and ask it to turn it into Super Brain notes.
• Edit directly. Open Super Brain to read, fix or archive any note.

How does shared memory change daily work?

An example scenario: your support agent learns from a customer chat that a new product ships in 5 days, not 3. It saves the decision to team memory. Within minutes:

• your inbox assistant tells email customers 5 days;
• your sales agent's next follow-up mentions the right date;
• your content writer stops promising 3-day delivery in posts.

Nobody had to tell each agent. (Your website chat bubble answers from its own help center, so ask an agent to update that article too.) That's the difference between several AI tools and one AI team. See how teams are set up in how to set up your first AI team.

Frequently asked questions

Can AI agents share memory with each other?

Yes, in KutrimBots. Team memory in Super Brain is open to every agent. Each agent's personal memory stays private unless you allow another agent to read it, just once, for 24 hours or always.

Is there a limit to how much Super Brain can store?

No. Super Brain has no size limit, because agents search it when needed instead of loading it all into every prompt.

Is my business knowledge used to train AI models?

No. Your workspace is kept apart from every other customer, and your data is never sold or used to train AI models. You can edit or delete any note at any time. See our security page.

What's the difference between Super Brain and a normal knowledge base?

A normal knowledge base is documents you search. Super Brain also stores the people, companies and projects in your notes and how they connect, keeps old facts as history, separates team and private memory, and is used by your agents directly while they work.

Does Super Brain understand Hinglish and typos?

Yes. Search matches words, close spellings and meaning, so "delivry charges pune" still finds your delivery note.

Sources

• A Definition of AGI (Hendrycks et al., arXiv)