Connect your assistant

Setting it up

Do step one once. Steps two and three take a minute each, and you repeat them for every assistant you want reading the same project.

Create your account

Sign up at withosmos.com with Google, GitHub, or an email address and password. Free covers your first project, so nothing is asked for up front.

If you'd rather start from inside Claude or ChatGPT, skip ahead — the connector's sign-in screen has a link to create the account on the way through, and you land back where you started.

Connect Claude

Settings → Connectors → Add custom connector. Name it Osmos and give it this address:

connector URL

https://withosmos.com/mcp

Claude opens the Osmos sign-in, you approve, and it's connected. Enable it in the project or chat you want it in, and set the Osmos tools to "Allow always" so you aren't asked every turn.

Connect ChatGPT

Settings → Apps → Advanced → Developer mode, then add a connector with the same address, https://withosmos.com/mcp, using its sign-in option. Approve, and it's connected.

Developer mode is where OpenAI keeps custom connectors today, and it needs a paid ChatGPT plan.

Ask for a context, then work normally

Say "use my Northwind API context" — or "make one" if you haven't yet — and carry on with whatever you were doing. From then on your assistant reads what the project knows before it answers, and keeps what's worth keeping afterwards.

You don't have to remember to do this. Most people say it once per assistant and never think about it again.

Bring in your team

Share the project, not your login

Open the context and press Share. Invite people by email at view, edit, or manage — or turn on "anyone with the link" when an account isn't worth the friction.

They connect their own Claude or ChatGPT the same way you did, in their own account — nothing to install on their side either. Their private notes stay theirs; what they publish to the project, everyone in it can use.

The moment it clicks

Your first handoff

Two people, two different assistants, nothing pasted between them. This takes about a minute and is the fastest way to know it's working.

Say something durable in ChatGPT

"We're using Supabase Auth with passkeys." It gets published to the project. Check the context's Memory tab — the decision should be there, with your name on it.

Ask Claude something related

From the second account: "Implement Windows auth." It already knows about the passkeys decision, with no prompting from you, and works from it.

Look at what it was shown

Open What AI saw and find that answer. It lists what went into it and counts anything held back — the receipt behind the result.

Check that private stays private

Have one person save a private note and confirm the other never receives it. Then change someone to view-only and watch their next attempt to publish get refused.

If something doesn't connect

The connector was added but the assistant never uses it

It has to be enabled in the specific project or chat, not only added in settings. Set the Osmos tools to "Allow always" while you're there, and ask it which context it is working in — it will tell you, or ask you to pick one.

It says it can't tell which context to use

That's deliberate. When more than one context is available and nothing says which, Osmos asks rather than guessing — a wrong guess would mix two projects' memory. Name the one you want and it stays there.

Someone can read a context you didn't expect

Open Share: it lists everyone who can read it right now, including invitations nobody has accepted yet and whether the link is on. Turning the link off stops new people joining and doesn't remove anyone already in.

You want an assistant to stop having access

Every assistant you've connected is listed on its own, and you can disconnect any one of them without touching the others. The next thing it tries to read fails.

One project understanding.

Across every assistant your team already uses.

Start free What stays private

Something not connecting, or a question before you start? Talk to us.