Team Workspaces: Roles, Sharing, and Shared Context

A team's AI problem is context scatter: everyone has their own chats, their own files, their own half-taught assistant. A workspace is the fix: one boundary holding the team's agents, files, tasks, and history, with roles deciding who can do what.

The model

Everything in idapt belongs to exactly one workspace. Your personal workspace is yours alone; create a team workspace and invite members with a role:

Owner / admin: manage members, settings, and governance.
Editor: full daily work: chats, files, agents, tasks.
Viewer: read access: the auditor and stakeholder role.

The workspace is the unit of sharing: put the project there and the team has it, without per-file ceremony (per-item sharing still exists for the exceptions).

Shared agents: the team's editing agent has one memory and one set of instructions: everyone gets the same behavior, and improving it improves it for all.
Shared Drive: sources, drafts, and deliverables in one tree; any chat can reference them.
Shared tasks: assigned work with boards and roll-ups, where agents are ordinary teammates.
Shared skills: house procedures every agent follows, versioned in the workspace.

Governance that scales past trust

Per-workspace feature toggles and an autonomy ceiling no chat may exceed.
Gateway governance: allowed models and providers per workspace, so "we only use these models for client data" is a setting, not a memo.
Run traces on agent work: who asked for what, what it did, what it cost.

Good to know

Members see workspace content per their role; personal workspaces stay invisible to everyone else.
Organizations layer billing and seats on top for larger teams; the workspace stays the working unit.
The workspaces feature page and help article cover setup and roles in detail.

Start with one project workspace and two teammates: move that project's files in, share one agent, and the context-scatter difference shows up in the first week. Consultants and analysts run whole engagements this way.

The model

Everything in idapt belongs to exactly one workspace. Your personal workspace is yours alone; create a team workspace and invite members with a role:

Owner / admin: manage members, settings, and governance.
Editor: full daily work: chats, files, agents, tasks.
Viewer: read access: the auditor and stakeholder role.

The workspace is the unit of sharing: put the project there and the team has it, without per-file ceremony (per-item sharing still exists for the exceptions).

Shared agents: the team's editing agent has one memory and one set of instructions: everyone gets the same behavior, and improving it improves it for all.
Shared Drive: sources, drafts, and deliverables in one tree; any chat can reference them.
Shared tasks: assigned work with boards and roll-ups, where agents are ordinary teammates.
Shared skills: house procedures every agent follows, versioned in the workspace.

Governance that scales past trust

Per-workspace feature toggles and an autonomy ceiling no chat may exceed.
Gateway governance: allowed models and providers per workspace, so "we only use these models for client data" is a setting, not a memo.
Run traces on agent work: who asked for what, what it did, what it cost.

Good to know

Members see workspace content per their role; personal workspaces stay invisible to everyone else.
Organizations layer billing and seats on top for larger teams; the workspace stays the working unit.
The workspaces feature page and help article cover setup and roles in detail.