What is shared memory for AI agents?

Shared memory for AI agents is a single store of facts that every agent on a team can read from and write to, instead of each agent keeping its own notes. It lets an agent know what the whole team knows: what was decided in a meeting, what a customer said in an email, who is meeting whom next week.

Why agents forget

An AI agent only knows what is in its current conversation, plus whatever its tool gives it. Close the chat and the context is gone. Some agents keep personal notes, but those notes belong to one person and one tool.

On a team, that breaks down fast:

• Your agent never heard the call your colleague took yesterday.
• A decision made in a meeting lives in someone's head, or in a transcript nobody reads.
• Each person re-explains the same background to their own agent, every day.

The agents are capable. They are just working from different, partial pictures of the company.

What shared memory means for a team

Most of a team's knowledge is not written down in a wiki. It sits in three places: meetings, email, and calendars. Shared memory turns those into facts about people, companies, and work, such as "the customer asked for a Q4 start" or "Ana owns the renewal", and keeps the source next to each fact: the quote, the date, and where it came from.

Two things make this different from a pile of transcripts:

1. Facts, not files. A search for a customer returns what is known about them, not forty documents to read.
2. Shared by default. What one teammate hears becomes context for every teammate's agent, with the right limits on who can see what.

What to look for

If you are comparing tools, four questions separate useful shared memory from a shared junk drawer.

Does every fact cite its source?

An agent that says "the customer wants a Q4 start" is only useful if you can check it. Look for a quote, a date, and a link back to the meeting or message. Without sources, you cannot tell a real decision from a guess.

Is access scoped per person and project?

A team brain that shows everyone everything is a liability. Each agent should see only what its teammate is allowed to see. Private email, for example, should stay private to its owner, with the team getting only the facts drawn from it.

Does it work with the agents you already use?

Your team will not switch agents for a memory tool. Look for support for MCP, the open standard agents use to connect to tools, so Claude, ChatGPT, Cursor, Codex and others can all use the same memory. Read and write both matter: agents should be able to add what they learn, not only look things up.

Is upkeep low?

Memory that someone has to curate by hand goes stale. The best setups fill themselves from the work that already happens, such as meetings, email, and calendar, and ask people for almost nothing.

How Nysa does it

Nysa is shared memory for teams that work with AI agents. It builds one company brain from three sources:

• Meetings. A notetaker joins the calls you choose, as a visible participant, and transcribes who said what.
• Email. Gmail history is imported when you connect, and new mail keeps flowing in.
• Calendar. Google Calendar events and attendees stay in sync.

Nysa turns these into facts with sources. Every teammate's agent reads from the same brain over MCP, and agents can also write back to it, each signed in as its teammate. Each agent sees only what its teammate can see; raw email stays private to its owner. Nysa works with Claude, ChatGPT, Cursor, Codex, OpenClaw, and Hermes, and setup takes under 5 minutes. Nysa supports Google accounts today.

FAQ