Connectors
First-class sources feed the same engines through one connector framework. Each connector surfaces both resources (files, sites, channels, repos, buckets) and identities (people, groups and guests — and the non-human agents, bots, apps and roles).
Scopes are read-oriented and least-privilege: the default is read-only, enough to map permissions and never more. Write scopes exist only for Active Prevention and auto-remediation, and are opt-in per connector. Every connector connects to live data today, and every connector can also be explored in demo (mock) mode.
Google Workspace
Surfaces Drive files and folders with sharing links normalized to public, external or org-wide; and Workspace people, groups and service accounts — including AI Drive assistants governed as agents.
Read-oriented OAuth scopes map the permission graph; a Google OAuth web client activates real mode (it also lights up “Sign in with Google”).
Microsoft 365
Surfaces SharePoint Online and OneDrive sites, libraries and files, with sharing normalized (anonymous → public, organization → org-wide, guest → external); and Entra ID members, guests, groups and — the flagship — app registrations and service principals, mapped to service or AI-agent identities (for example, an over-privileged Copilot app).
Least-privilege Microsoft Graph read scopes first (Sites.Read.All, Files.Read.All, Directory.Read.All, Application.Read.All); write scopes are reserved for prevention and opt-in.
Slack
Surfaces workspaces, public and private channels, Slack Connect external shared channels, files with public links, and multi-channel guests; plus installed apps, bots and integrations with their OAuth scopes, governed as agents and services — an over-broad AI bot scope surfaces as critical.
Read scopes first via OAuth; write access sits behind the per-connector prevention opt-in.
Notion
Surfaces the workspace, teamspaces, pages and databases, with share-to-web, guest and external sharing; plus integrations and bots with their capabilities, governed as agents — a publicly shared sensitive page surfaces as critical.
Read scopes first via OAuth; write access sits behind the per-connector prevention opt-in.
GitHub
Works for a personal account or an organization — detected automatically from the account type. Organizations: surfaces the org, repositories (public, internal, private), outside collaborators and teams, plus GitHub Apps, OAuth apps, fine-grained PATs and deploy keys, governed as agents and services. Personal accounts: inventories your installed GitHub Apps and their permission grants, your owned repositories, per-repository deploy keys (including write-capable keys), and collaborators — the forgotten-access surface of an individual developer’s account.
Read scopes first via OAuth; write access sits behind the per-connector prevention opt-in.
AWS IAM
Surfaces the account, IAM users, roles, groups and policies, access keys, and S3 bucket and role-trust exposure; plus roles assumable by services or automation, governed as service and AI-agent identities. A wildcard (*:*) admin role, a public S3 bucket and cross-account trust surface as critical.
Connects with read-only IAM credentials or an assumed read role (api-key style) — enough to enumerate, never to change anything.
One model, every source
The framework generalizes: each new connector reuses the shared posture rules, blast radius, scoring, cross-scan diff, the verdict / Agent Guard engines, the compliance generator and the remediation pipeline — unchanged. A single compliance report can cover every connected source, and agent governance spans every vendor's AI agents.
App and role identities are workspace- or org-resident, so their risk surfaces through the AI-agent and service rule family — never miscounted as external collaborators.