Connect once. Every assistant catches up.

Osmos is a shared memory for a project. Your assistant reads it at the start of a task and writes what it learned back at the end, so the next conversation — in any assistant, by any teammate — starts from what is already known.

Setting it up

Two minutes, and the only thing you need to copy is one address.

1. Create an account Sign up at withosmos.com with Google, GitHub, or an email address and a password. Confirm your email when the message arrives — you can start straight away, but sharing a context with someone else waits until you have.
2. Add the connector In Claude, open Customize › Connectors, choose Add custom connector, and paste this address:
https://withosmos.com/mcp
In ChatGPT, turn on Developer mode under Settings › Apps › Advanced, then add the same address.
3. Approve the connection Your assistant sends you to Osmos to sign in. You will see exactly which account is being connected and what the assistant will be allowed to do, and nothing happens until you approve it. Each assistant you connect is listed separately and can be cut off on its own.
4. Work normally Ask your assistant what the project has already decided, or tell it to remember something. There are no commands to learn. If you have no context yet, say “make one” and it will.

Claude requires a Pro or Max plan to add custom connectors. ChatGPT’s developer mode is available on Plus, Pro, Business, Enterprise and Education plans, on the web.

What to ask it

Ordinary sentences. The assistant decides when to reach for Osmos.

“What have we already decided here?” Pulls the project’s decisions, constraints and anything still unresolved into the conversation, so you are not re-explaining context you have explained before.
“Remember that we chose Postgres, because we need full-text search.” Records a decision with its reason. Every other assistant and every member of that context now starts from it.
“What changed since I last worked on this?” The catch-up view — what teammates and their assistants added while you were away.
“Switch me to the mobile app project.” Contexts work like documents. You can keep several and move between them, and each assistant remembers which one it is in.

Sharing with people

Like sharing a document, with one deliberate difference.

Invite by email, or by link Give someone view or edit access. Their assistants pick the context up on the next sync — they do not have to do anything beyond joining.
Only people can share An assistant can read and write memory, and can never grant anyone access to anything. There is no tool for it to call. That is the difference: sharing is a decision a person makes.
Private stays private Anything you record privately is yours. It is excluded when the shared context is built for someone else’s assistant — not filtered out of the answer afterwards, but never put in front of the model at all.

What it will not do

Worth knowing before you rely on it.

Send your memory anywhere Finding what is relevant to a task happens inside the same database your memory lives in. There is no embedding model and no summarising model, so no third party — including any model provider — ever receives the content. Nothing is used to train anything.
Delete what you recorded Correcting a memory marks the old version replaced rather than removing it, and both stay readable. Nothing is ever deleted to enforce a plan limit either — going over a limit stops you adding more, it does not destroy what is there.
Read your files or your screen Osmos only holds what you or your assistant deliberately record. It has no access to your machine, your repository, or anything else your assistant can see.

Questions, or a request to see or delete what we hold: hello@withosmos.com. What we store and for how long is set out in the privacy policy.