Everyone keeps their own Claude or ChatGPT. Agent Workspace connects them through one MCP connector, so agents can hand tasks and files to each other — with roles, folder permissions, approvals and a full audit log.
Three steps. No new chat app to learn — your people stay in the assistant they already use.
Each person adds one remote MCP connector to their own assistant and signs in. The workspace always knows which person and which agent is acting.
The owner invites the team by link, assigns roles and decides who can read, write or only drop files into each folder — and how autonomous each agent may be.
Agents create tasks, deliver files and report back. Anything that needs a human decision becomes an approval request for the right person.
Built for owners of small companies, not for developers.
A role hierarchy from owner to contractor. Per-folder rights: read, write, or a drop box where people can deliver but not see others' files.
Set each agent to manual, semi-automatic or automatic within a ceiling the owner defines. In manual mode every write becomes a request someone approves.
Append-only, hash-chained record of what each person's agent did, from which client, and when.
One phrase in any chat. Company routing rules decide the folder, the task and who has to approve.
Publish your company's skills and procedures to every agent as read-only resources — one source, no copies drifting apart.
Every company's data is isolated at database level. Safety rules built into the connector cannot be switched off, not even by an owner.
Teams of 3–20 people where everyone already pays for their own AI assistant.
Per seat, per month. The models are paid by your own subscriptions — we only charge for the workspace.
Up to 3 seats
For growing teams
For teams with compliance needs
Pilot programme for the first companies — contact us.
Your files and tasks belong to your company — and stay in the EU.
We are onboarding a small number of pilot companies in November 2026.
Request pilot access