Agent routing that respects your identity boundary.
Give agents and workflows access to current organizational context without handing them unrestricted directory credentials or embedding identity logic in every application.
Roster separates participant resolution from authentication, authorization, source directories, and workflow execution — so security controls stay where they belong.
Every resolve call flows through five governed layers
An agent asking “who should act?” never touches the directory directly. The request passes five independent checks before Roster returns a participant — each one owned by a different security concern, each one auditable on its own.
One identity model for humans, agents, and workloads
Every actor gets its own Roster identity — with distinct authentication, distinct credentials, distinct lifecycle, and distinct audit attribution. No shared keys, no ambiguous callers.
Authenticate through your configured login providers and operate under their mapped Roster identity — with full audit trail tied to the corporate SSO subject.
First-class identities for autonomous agents, with dedicated API keys, ownership metadata, and per-agent scopes so a compromised agent never impersonates a human.
For applications, integrations, scheduled jobs, and workload automation — long-lived where needed, revocable at any moment, never confused with a person.
Belong to exactly one identity, carry explicit scopes, rotate independently, and are revocable without disturbing the identity or its history.
Directory-backed. Not another IAM system.
Roster reads organizational data from the directories you already run. Providers stay the source of truth. Roster materializes only the records and membership relationships needed for resolution — never the whole workforce, never for every agent.
Credentials with surgical scope, not blanket access
A Resolve-only agent needs four scopes, not thirty. Every API key is issued to a specific identity, carries an explicit capability list, and can be revoked or rotated without touching the rest of the platform.
Scope permits the call. The acting identity still needs access to the underlying project, participants, and records. Even an over-permissive key can only reach what its identity is authorized to see.
Authorization is not resolution — and vice versa
Policy engines answer whether a known actor may perform a known action. Roster answers who the actor should even be. Both belong in a production workflow. Neither replaces the other.
OPA, OpenFGA, Cedar, or your IAM system evaluates permissions on already-known subjects. Roster hands them a governed candidate to evaluate.
For any workflow context, Roster returns the user, group, role, or delegate that current organizational state points to — then your authorization layer takes it from there.
Roster resolves the candidate. Authorization verifies the candidate may act. The workflow engine executes. Three separable concerns, three separable audit trails, one accountable outcome.
Security controls built into the resolution path
Every control below is enforced server-side, at the moment of resolution — not left to prompt engineering, agent code, or workflow authors.