Answers
Shared memory for teams that work with AI
AI memory for teams is memory that the whole team and its AIs read, not memory that belongs to one person's chat window: the team's decisions, rules and daily record, written down outside every assistant, with who decided each one and whether it still holds. People use three things for this today: a shared wiki or workspace such as Notion, instruction files in the repository, and the memory each AI tool keeps for the person using it. Arroway is the memory the team's AIs read before they start a task and write to when they finish. The memory belongs to the project, so the second person's assistant starts from what the first person already settled, instead of from a blank. It works the same with one person and one agent, because the next session is a reader too, and it installs in the tool you already use: Claude, ChatGPT or Codex, Cursor, or anything that connects over MCP.
Last updated October 27, 2026
What Claude's own memory announcement says about teams
On 25 August 2026 Anthropic announced that the memory Claude uses in chat is the same one it uses in Claude Cowork, kept as a list of short topic files you can read, edit and delete. The scope that announcement declares is one person and one account: on Free, Pro and Max it is on by default, and on Team and Enterprise an admin makes it available while each individual still has to turn it on. It says nothing about memory scoped to an organisation, shared between people, or carrying authorship — the text addresses one reader throughout. Arroway is built for the part of the question that is left over: what the team decided, whether it still stands, who wrote it, and the rule that a memory survives its author leaving the project.
What people use today as team memory, and what it leaves open
A shared wiki or workspace: Notion's MCP guide, read on 29 September 2026, says that after you authorize the connection the client can read and update the Notion content you can access. Instruction files in the repository: Cursor, GitHub Copilot and Claude Code each read their own file or files from the project, as their documentation describes, so they reach the agents working in that repository. The memory inside each AI tool: the one Anthropic announced on 25 August 2026 is scoped to one person and one account, as the note below says. Each of them is somewhere a person can write, and each answers a part of the question. The team's problem is different: the same rule gets explained again in every conversation, in every tool, by whoever happens to be there — and the one time nobody explains it, the assistant guesses. Arroway keeps decisions, rules, references and a dated log as material of the project, with the name of the person who wrote each line, so what was settled on Tuesday is still there on Thursday, for someone else, in a different tool.
The AI reads first, then works — not the other way round
When a task starts, the assistant asks Arroway for what applies: the rules that always hold, the decisions closest to this work, and what teammates have been doing lately. Then it works. When it finishes, it writes back what the work settled — not the conversation, just what someone arriving next week would need. The read also says how much came back and what did not fit, so “the AI has the rule” is something you can check rather than hope.
Every line has a name on it, and it stays when the person leaves
Each memory and each log entry says who wrote it — a person answers for what their assistant recorded. Taking someone off a project removes their access and nothing else: what they wrote stays, with their name, still served to everyone who remains. What a project remembers belongs to the project, not to whoever typed it.
What it looks like in practice
On Monday, one person settles how refunds are handled, and their assistant records the rule with the reason. On Tuesday a teammate opens a fresh conversation in a different tool and asks how to handle a refund. Their assistant reads the project first and answers with Monday's rule — nobody re-explained it, nobody was asked, and nobody needed to remember that a rule existed.
Questions people ask about this
- Is this the same as the memory built into my AI?
- They are not the same thing, and Arroway does not publish a comparison it has not verified. What Arroway is: a place the team keeps, where a person approves what governs, and which every member's AI reads. Where a vendor states the scope of its own memory, this page cites that statement and links to it — the note above is Anthropic's announcement of 25 August 2026, which describes memory for one person and one account. How the two behave alongside each other in practice is not something this page will assert for you.
- Does everyone need the same AI tool?
- No. The memory lives on a server that each person's assistant connects to, so someone in one tool and someone in another read and write the same project. Nothing is copied between them; there is one record.
- What happens the first time, when nothing is recorded yet?
- It starts empty, on purpose. The first useful thing to do is point your assistant at the documents, rules and decisions that already govern the work; it proposes what to record, and you keep what belongs and throw out what does not. You can also write a rule yourself, without waiting for an AI to notice it.
Where this is verifiable
Product documentation on this site (How it works, Install) and the sanctioned product spec in the repository. The claim about departures is a standing project rule, not an inference. The note on native memory repeats only what Anthropic's announcement of 25 August 2026 declares about itself, and links to it; this page does not assert how that memory and Arroway behave alongside each other. Notion MCP (developers.notion.com/guides/mcp/get-started-with-mcp) is described only as that page described it on 29 September 2026, and the instruction files as the documentation of Cursor (cursor.com/docs/context/rules), GitHub Copilot (docs.github.com/en/copilot) and Claude Code describes them, read on the dates recorded on the answer about alternatives to CLAUDE.md.
https://www.arroway.app/en/answers/shared-memory-for-ai-teams