Solutions / Private AI workspace

A collaborative workspace with AI and access control.

A private AI workspace is a place where a team works with AI without handing the AI everything. In SweetHive, people, apps and AI agents share one structure of contexts and groups: each agent is personal and reads only what its owner can see, the model can run on your own hardware, and the workspace is hosted on AWS in Ireland.

Who it is for

Teams that want AI on their real work, not on a copy of it.

It fits small and mid-sized organizations that keep client work, internal matters and external partners in the same tool, and need the AI to respect those separations without a dedicated security team configuring it.

How it differs from a general AI chat

The AI does not get its own view of your data.

A general AI chat

You paste or upload what you want it to read, or you connect it to a whole drive. What it can reach is decided connection by connection, separately from who is asking.

An AI layer with its own index

The assistant indexes the company's content and filters answers by permission at query time. It works, and it depends on every permission being right everywhere.

SweetHive

There is no separate AI index. An agent belongs to one person and asks the workspace for content with that person's visibility. If the person cannot open it, the agent cannot read it.

Access

One structure decides what people, apps and agents see.

You do not maintain a second set of permissions for the AI.

Hives and contextsA hive is your organization. Inside it, contexts are nested like your work: company, department, project, client.
GroupsPeople see a context because a group they belong to is active on it. Visibility flows down the tree, and nothing else grants it.
AgentsEach member connects personal agents and chooses their scope: all their hives, one hive or one context. The agent's reach is that scope intersected with what the member can see.
AppsAn app installed in a context works inside that context and the sub-contexts the user's groups reach.
External peopleGuests are added to a group on a specific context and see that only. More on external collaboration.
Where things run

You choose where the model runs.

The modelOn each person's computer with the free Agents Node app, or on a server of your organization with Agents Server (Organization plans). More on local AI agents.
The workspaceThe application, the database and the files are hosted on AWS in Ireland (eu-west-1). Transactional email is sent through a service located in the United States.
Cloud AIThe SweetHive platform does not send your content to an AI provider. A person can choose to use their own Claude account in Agents Node; then their prompts are processed by Anthropic. Apps from the Store may use cloud AI services of their own.
Limits and prerequisites

What it does not do today.

SearchAgents find content by keyword in messages and read attachments up to 20 MB (PDFs, scans, images, text). There is no semantic search across the whole company, and Office files are not read.
Activity recordsFor each agent you see when it was last used and how many items it read, and you can revoke it at any time. There is no consultable audit log per context yet.
ConnectorsGmail and Google Calendar in read-only mode, and Google Drive by picking single files. Other connectors are not available yet.
DeploymentThe workspace is a cloud service. Only the AI can run on your own hardware.
CertificationsSweetHive holds no security certifications today. The security page lists infrastructure and subprocessors.
Example scenario

One question, three different answers.

This is an illustration, not a customer story. In a consulting firm, three people ask their agent "what is new on the Rossi account?". The partner's agent answers from the client context and the internal pricing context. The consultant's agent answers from the client context only. The client's own contact, invited as a guest, has no access to the internal contexts at all, so nothing from them can appear anywhere on their side.

how
FAQ
What does "private AI" mean here?

Two things: the AI reads only what the person using it is allowed to see, and you can run the model on hardware you control instead of a cloud AI service. It does not mean the whole product is installed on your premises.

What data can an agent read?

The messages and attachments of the contexts in its scope, limited to what its owner can see at that moment. If the owner loses access to a context, so does the agent.

Can I choose where the model runs?

Yes: on each person's computer with Agents Node, or on your organization's server with Agents Server on Organization plans. An administrator can require the company server for everyone.

How are external users handled?

They are invited as guests into a group on a specific context and see only that. Guests are free, within the number included in your plan.

How much does it cost?

Plans are priced per team, not per user. Free covers 3 members; Pro is € 29 + VAT per month for up to 10 members; Organization plans start at € 49 + VAT per month. Every paid plan starts with a 14-day trial.

Read more: Local AI agents · External collaboration · Agents · Security · Pricing · Groups in the documentation · Connecting an agent

See it on your own structure.

Create a hive, add two contexts and connect an agent: in a few minutes you can check what it can and cannot read.

Try SweetHive free for 14 days