How to share AI projects with your team — share the room, not the chat

Open the forums where people actually run AI at work — the OpenAI community, the Claude subreddits, the YC threads — and the same sentence keeps surfacing in different words: I want to share the project, not the chats. People do not want to forward a transcript or paste a read-only link. They want to hand a teammate the whole working context — the instructions, the files, the accumulated memory, the back-and-forth so far — and have that teammate pick up exactly where the work is.

That is a precise request, and most AI tools answer a different one. They let you share a conversation. The unit teams need to share is a project. This piece is about why those are not the same thing, and what changes when the thing you share is a room rather than a chat.

Why "share this chat" never feels like collaboration

A shared chat link is a snapshot. Someone clicks it and sees what was said up to that moment, usually read-only, frozen. They cannot add a turn that the original author will see. They cannot bring the model the same instructions and files the author had. They are looking through a window at someone else's work, not standing in the room.

This is why a recurring complaint about team AI tiers is so blunt: people say the "Team" tier feels less collaborative than two separate accounts, that the name promised something the product did not deliver. The frustration is real and it is not about any one vendor being lazy. It comes from the underlying shape. When a product is built around a private 1:1 thread, "sharing" can only ever be bolted on as an export. The collaboration the team wanted — two people and an AI working the same live context — was never the primitive.

To be fair to the incumbents, they have moved. ChatGPT's Projects and Group Chats and Claude's Projects all let a team gather documents and context in one place, and that is genuinely better than emailing prompts around. But there is a difference between a shared folder the model reads and a shared space the team and the model both live in. The first is storage. The second is a workspace. SquidHub is built as the second.

The unit to share is a room

On SquidHub the thing you invite someone into is a room, and a room carries three things at once that a chat link cannot:

• The full transcript, live. Everyone in the room sees the same thread and can add to it. Your teammate is not reading a frozen copy — they are writing the next turn, and you see it. This is the multiplayer mode that ordinary AI chat is missing: more than one human, present at the same time.
• Persistent room memory shared by everyone. A room accumulates facts — decisions made, constraints agreed, the shape of the thing you are building — and every member and every AI in the room draws on the same memory on every turn. This is the part that breaks the duct-tape stacks people build out of Slack plus a sync job plus the API. You stop re-establishing context each session because the room remembers it for the whole group.
• The AI participants themselves. The squids in the room are team-usable. A teammate does not need to recreate your assistant from a screenshot of its instructions. The squid is already there, already configured, already carrying the room's history.

The AI stops being a private tool you each own a private copy of, and becomes shared infrastructure the team works through together.

That last point is worth slowing down on, because it is the one most teams have been quietly fighting.

A squid is a team member, not your personal GPT

A custom GPT can only be edited by the person who made it. People run straight into this: unable to collaborate on a custom GPT. You built a useful assistant, your colleague wants to tweak its instructions, and they cannot — it belongs to your account, not to the team. So they rebuild it, slightly differently, and now there are two assistants drifting apart.

A squid belongs to the workspace, not to one login. It is built once in the /new-squid wizard from structured fields — its occupation, its traits, its instructions, the knowledge it carries — and then it is a member of the room like any person. Several humans can work with the same squid. A teammate can bring their squid into your room and the two can disagree about the right call while you both watch. The AI stops being a private tool you each own a private copy of, and becomes shared infrastructure the team works through together. We make that argument in full in multi-agent collaboration.

Priced for two, not five

The other reason "Team" tiers disappoint is the bill. The common shape of a collaboration plan is a seat minimum — you cannot buy collaboration for two people; you buy it for five, whether you have five or not, and the math against two individual plans rarely works out. So people do the rational thing: they make a second login, share an account, and route around the product. That is not a pricing edge case. It is a signal that the smallest real team — two people — has nowhere to land.

SquidHub is built for that team. You can share a workspace with one other person without buying three empty seats to unlock it, and you can pull someone into a single room as a guest — Slack-style, single-channel — when you only need their eyes on one thing. A guest sees the rooms they were added to and nothing else: not your other rooms, not your squids, not the member roster of the workspace. The guest model is the "second login to avoid sharing full account access" instinct, done properly, so nobody has to hand over the keys to everything to collaborate on one thing. If you want the workspace-and-room model laid out end to end, the docs walk through it.

Who sees whose data in a shared project

The moment a project has shared context, a fair question follows: who can see what? A skeptical team lead should ask it, and we would rather answer it plainly than wave at a badge.

Two boundaries do the work. The first is tenant isolation: rooms and squids belong to a workspace, and joining one room never grants access to the rest of the workspace — a guest added to a single room stays scoped to that room, enforced on the server, not just hidden in the UI. The second is encryption at rest: message text, room and user memory, squid instructions and uploaded files are stored as ciphertext in the database (AES-256-GCM), so a database dump or a curious operator browsing storage does not read your project.

We are equally clear about the edge of that promise. SquidHub is a hosted service that runs the AI for you, so it is not end-to-end encrypted — the live application must be able to decrypt content to function, and the AI provider sees the conversation transiently to answer. We do not claim otherwise. What we do hold is a no-training, zero-retention posture with our AI provider, a short and named list of operators with production access, and permanent deletion on request. The full account, including what is deliberately left in plaintext, lives in our security page and privacy page, with the trust posture summarized at the trust center.

Bring your own model, or use ours

Teams rarely standardize on one model anymore, and a shared room should not force them to. In a SquidHub room you can run squids on Anthropic, OpenAI, xAI, or Google Gemini, and you choose how the brain is paid for: bring your own provider key, in which case a turn costs nothing on our side and runs under your own account and contract, or use the managed SquidHub AI tier, metered in ink and free during the current beta. Different squids in the same room can run on different models. We cover the trade-offs of owning your keys in bring your own key.

FAQ

Can I share a project with one other person without buying a team plan

Yes. You can share a workspace with a single collaborator, or add someone to one room as a single-channel guest, without a five-seat minimum. The unit of collaboration is two people, not a five-seat block.

Does my teammate see the AI's instructions and memory, or just the messages

They get the working context — the live transcript, the shared room memory, and the squids configured in the room. It is not a read-only snapshot of a chat; it is the project itself, and they can add the next turn.

Can two people use the same AI assistant

Yes. A squid belongs to the workspace, not to one account, so several people can work with the same squid, and a teammate can bring their own squid into your room. There is no "only the owner can edit this assistant" wall.

Is the shared context encrypted, and can SquidHub read it

Content is encrypted at rest with AES-256-GCM, so a database leak yields ciphertext. It is not end-to-end encrypted: a hosted service that runs the AI must be able to decrypt to operate, and we say so plainly. We run a no-training, zero-retention posture with our AI provider and delete permanently on request.

Can a guest see the rest of my workspace

No. A guest sees only the rooms they were explicitly added to — not your other rooms, squids, or member list — and the scope is enforced on the server.

Share the project, not the chat

The pattern under all the forum complaints is one mismatch: people are asked to share a conversation when what they have is a project. Once the thing you share is a room — transcript, shared memory, and team-usable AI together — collaboration stops being an export feature and becomes the default state of the work. That is the difference between a share-link and a workspace, and it is the whole idea behind multiplayer mode for AI.

If you want to try sharing a project the way you actually meant it, open a room or book a demo. Questions about how shared context is isolated and stored go to hello@squidhub.ai.

— SquidHub Team