---
title: What Atmina is
description: Atmina is shared memory for your team's agents — capture what your team decides, and recall it later with the sources behind it.
owner: markus
last-reviewed: 2026-09-30
---

# What Atmina is

Atmina is shared memory for your team's agents. Connect an agent once, and it can capture what your team decides and recall it later, with the sources it used.

Most of what a team knows lives in scattered chats, documents, and someone's head. Atmina gives your agents one place to keep the decisions, findings, and notes worth remembering — and to bring them back when you ask.

## The problem it removes

Every session starts from nothing. You re-explain the same architecture, restate the same constraints, and re-argue decisions the team already settled, because the agent that heard them last week has no way to reach them this week. What context does survive is trapped in one tool, on one machine, in one person's history.

Atmina moves that context out of the harness and into a place every agent on the team can reach. What one person's agent learns, the next person's agent can recall — and can show you where it came from.

## Who it is for

Atmina is built for the person running AI tools for a team of five to fifty: a founder-CTO, a staff engineer, an embedded consultant. You work across several agents and harnesses, you need decisions to survive between sessions, and you need to see what an agent knows and where it got it.

Team owners are the second audience. Shared memory is only worth having if you can decide who reaches it and check what it has been used for.

## What an agent gets from it

- **Recall with sources.** Every answer traces back to the file it came from, so a claim can be checked instead of trusted.
- **Capture in place.** The agent writes what you decide straight into the [Knowledge Base](https://atmina.ai/docs/knowledge-bases), with no copying between tools.
- **The same memory everywhere.** Claude, Claude Code, Cursor, and any other MCP-compatible client reach the same base over one endpoint.
- **A record.** Every recall and capture is on the record — which agent, which person, what it touched, what came back.



Agents can also read this Guide and search it without an account. See [Atmina for agents](https://atmina.ai/docs/for-agents) for the addresses and what each one allows.

## What stays under your control

Your team's Knowledge Bases are stored and searched separately from every other team's, so one team's memory is never mixed into another team's recall. [Roles and sharing](https://atmina.ai/docs/sharing-and-roles) decide who can recall what. You can export a base, or erase it, when you are done with it — see [your data](https://atmina.ai/docs/your-data).

## The shape of it

- A [Knowledge Base](https://atmina.ai/docs/knowledge-bases) is where your team's shared memory lives.
- You [connect your agent](https://atmina.ai/docs/connect-your-agent) over MCP, then [capture and recall](https://atmina.ai/docs/capture-and-recall) in plain language.
- [Roles](https://atmina.ai/docs/sharing-and-roles) decide who can recall what.


Atmina is in alpha, built in Australia by a small team. See [what's in alpha](https://atmina.ai/docs/whats-in-alpha) for the current product boundary.