Spaces: where humans and agents do work together
A Space is the unit of trust in Aimable. It bundles who is allowed in, what knowledge they can reach, which AI they can use, and what rules apply. The same for humans at the Workbench and for AI agents calling over MCP.
Every Space has four layers
You compose a Space once. The Platform then enforces those four layers on every message, every tool call, and every agent run. No matter who or what is driving.
Membership
whoNamed humans and named agents. Same directory, same roles, same revocation flow. One mental model.
Collections
whatCurated knowledge the Space can reach, and nothing it can't. Versioned, reviewable, promotable.
Policies
howPersonal-data rules, redaction, approval gates, model allow-lists, data-residency. Enforced at runtime.
Models
withThe approved model pool for this Space. Routed by intent, cost, and policy. Never a free-for-all.
One contract. Humans and agents included.
Same rules for everyone
Humans in the Workbench and agents over MCP/API share the same membership, the same policies, and the same logbook.
Blast radius is a boundary
A compromised account or rogue prompt is contained to its Space. Not a company-wide incident.
Built for regulated work
Spaces map cleanly to audit scopes like a case team, a product squad, or a client engagement.
Examples from real deployments
Claims handling
Claim handlers, adjusters, and an automation agent share one Space. Same policies, same logs.
Deal team
Analysts, partners, and a memo-drafting agent work inside a single Space per portfolio company.
Legal review
Counsel and a diligence agent collaborate on a mandate with jurisdiction-locked models.
Engineering squad
Code review agents operate in a Space with read-only repo access and blocked network egress.
“We stopped explaining "does this go to the cloud LLM". A Space already answered it.”
See what a Space looks like end-to-end.
Thirty minutes. One Space, walk it top to bottom. Members, collections, policies, and an agent running inside it.