Skip to main content
Smaya is Exterview’s in-app AI companion. It opens from a persistent Ask control in the header as a docked side panel that reflows the workspace rather than covering it, so it reads as part of Exterview, not a popup. It understands what’s on your screen, answers grounded in your organization’s knowledge with citations, and is built to grow, under human control, from answering into acting.
Smaya is one companion delivered as three tiers of escalating autonomy on a single governed foundation. What ships first is deliberately narrow and read-only. Every tier beyond that acts only with a human approving each step. Smaya never makes an autonomous adverse decision.

​
The three tiers

The foundation stays the same at every tier: grounded and cited, specific to your organization, human-approved, and audited. Each tier adds capability on top of that line, it never loosens it. One governed foundation carries three tiers of increasing autonomy. Tier 1, read-only, ships first; Tiers 2 and 3 are on the roadmap and never loosen the foundation’s controls.

Simple: read-only Q&A

Available first. Grounded, cited, state-aware answers about the record on screen and your organization’s knowledge. Smaya knows which candidate, job, and organization you are viewing, and says “I don’t know” rather than inventing an answer. No data ever changes.

Super: agentic co-pilot

On the roadmap. Smaya reasons, plans a multi-step approach, acts, and observes the result in a loop. Every step that changes data passes through a human approval gate first. Nothing consequential executes on its own.

MCP: external surface

On the roadmap. Reach Exterview from approved external assistants, such as Microsoft 365 Copilot, through a governed gateway scoped to your organization’s access rules. See MCP.

​
The line that does not move

These controls hold identically at every tier. This is what makes escalating autonomy safe to adopt.

Grounded and cited

Every answer traces to a source. Empty or low-confidence retrieval returns “I don’t know,” never an invented answer.

Specific to your organization

Every retrieval is filtered to your organization, with no access to another customer’s data, and sensitive fields are redacted by the viewer’s role.

Human-approved actions

Read and navigate assist you directly. Anything that changes data is proposed, not performed, and waits for a person to approve it. See Human Oversight.

Fully audited

Every answer, proposed action, and approval is recorded with its trace, so a reviewer can see exactly what Smaya did and why.

​
Where to go next

How It Works

The agent loop behind the tiers, and how Smaya decides it is done.

Human Oversight

The approval gate and the control model, from the first release onward.

​
FAQs

The first release is Tier 1: read-only, grounded, cited answers. Tier 2 (acting) and Tier 3 (external reach) are on the roadmap and add autonomy in stages, always with a human approving each action that changes data.
No. Read and navigate assist you directly. Any action that changes data is proposed and held at a human approval gate. Smaya never executes a consequential action autonomously.
From your organization’s indexed knowledge and the record on screen, specific to your organization. Every answer carries a citation back to its source.
Tier 1 is on the near-term roadmap. See Upcoming for timing, and treat Tier 2 and Tier 3 as later steps on the same foundation.

How It Works

The perceive, reason, act, and observe loop.

Responsible AI

How human review and no autonomous adverse decisions are enforced.