Skip to main content

​
Purpose

Smaya separates what every workload needs from what only one workload needs. The platform holds identity, permissions, approvals, evidence, integration and evaluation. Each application holds the expertise of its business function. That separation is why the second workload is far cheaper than the first.

​
The stack

Solid lines are Available today. Dotted lines are Upcoming.

​
What the Agent OS provides

Shared by every application, and never rebuilt per workload.

Identity and authority

Who an agent is, what it may reach, and the ceiling on what it may do. Authority is always the intersection of the agent’s own scope and the person who invoked it — never the sum.

Approvals and human control

Which actions run directly, which are proposed and wait for a person, and who that named person is. Consequential actions are always human-approved.

Evidence and audit

A permanent record of every consequential decision and every approval, reproducible after the fact. See Governance & Control.

Integration

Connections to the systems of record an organization already runs, and a governed way for approved external assistants to reach in. See Microsoft & Interoperability.

Lifecycle

Versioned agent skills, and job criteria locked once candidates are in flight. Upcoming: staged rollout and two-person change control for every AI Employee. See Agent Studio.

Evaluation

Automated checks on every scored output, with results in Scoring Logs. Upcoming: continuous fairness measurement that blocks a degrading change before release.

​
What an application provides

An application is the domain expertise: the vocabulary, the criteria, the evidence standards and the regulatory obligations of one business function.

​
What an agent is

An agent is a specialized worker accountable for one job. It is pre-built, not assembled by the customer. Every agent declares five things before it may run: Not every agent uses a language model. Some are pure rules and connections, and they are labelled as such.

​
What an AI Employee is

An AI Employee is the deployable unit. It is a named role, staffed with one or more agents, carrying a charter and a named human owner. Your teams work with AI Employees; they do not operate raw agents. See AI Employees.

​
How work runs: the agent swarm

Upcoming. For a role, the pieces above will combine into an agent swarm: one run per role and one sub-run per candidate, drawing on agents from the library. The swarm does the work between reports, and people answer reports and schedules in the Agent Inbox. Today, screening through to a ranked Leaderboard is Available. See The Agent Swarm.

​
Where Smaya’s assistant fits

The conversational surface — asking questions, getting cited answers, proposing actions — is one surface of the OS, not the whole of it. It is Available in Exterview and in connected assistants over MCP. Upcoming: read-only asking from Slack. See Smaya Assistant.

​
FAQs

Because they answer different questions. An agent answers “what can this do?” An AI Employee answers “who is accountable for this work, and what are they allowed to do?” Organizations manage roles, not capabilities.
Yes. Agents are reusable building blocks. The same Resume Screening Agent can be part of several AI Employees, each with different authority and a different owner.
No. Applications inherit the OS controls and may narrow them, never widen them.