Governance & Access
MindGraph controls what each human or agent can know and what each agent can do. Identity, knowledge scope, tool grants, policies, approvals, and execution records are evaluated as one authorization path.
The goal is least-privilege context: an agent should retrieve only the knowledge it needs, invoke only the tools it was granted, and stop for approval when policy requires judgment.
Authorization path
- Resolve identity. The server derives the human or agent principal from the credential.
- Resolve knowledge scope. Organization membership, Space grants, Project roles, and classification ceilings determine readable context.
- Check the tool grant. The agent must hold an explicit grant for the native or MCP affordance.
- Evaluate policy. Published policies allow, deny, or require approval based on the agent, tool, arguments, and context.
- Execute and record. The decision, approval when present, execution, and resulting mutations are retained for audit.
Knowledge scope
| Boundary | What it controls |
|---|---|
| Organization | Workspace membership and administrative role. |
| Space | Read or write access to a knowledge partition, with an optional classification ceiling. |
| Project | Viewer, editor, or agent authority over a source-bounded working context. |
| Tool | Permission for a specific agent to invoke a registered native or MCP affordance. |
A source's PartOfProject relation defines corpus membership. It does not grant access. Project roles are represented separately so admission and authorization cannot be confused.
Spaces
Spaces partition knowledge inside an organization. A grant gives a principal read or write access; a classification ceiling can further restrict which labels that principal may see. Owners and admins manage grants.
Open an agent's Governance tab to review and change its Space access and classification ceiling.
Projects
Project roles limit retrieval and mutation to an admitted source corpus. See Projects & Living Briefs for the source boundary and role semantics.
Project scope is selected when an owner or admin configures an agent. The dashboard creates the corresponding Project role and prevents an API key from changing that authority.
Policies
Structured policies can be tested in shadow mode, then published for enforcement. The organization chooses permit-by-default or deny-by-default behavior; published policy decisions are synchronized into the graph so the decision and its provenance remain queryable.
Owners and admins manage policy mode, drafts, publication, and synchronization from an agent's Governance tab.
Approvals and audit
When a policy requires approval, the tool call is parked with its arguments. An authorized dashboard user can allow or deny it. Before an allowed call resumes, MindGraph checks that access, the tool registry, and policy inputs have not drifted.
Pending MCP approvals are resolved in Integrations; the agent Governance tab shows its policy and execution history.