> ## Documentation Index
> Fetch the complete documentation index at: https://docs.exterview.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# The Agent Swarm

> Upcoming: give Smaya the role, and it builds the pipeline, talks to every candidate, gathers the evidence, and brings you the people worth deciding on. What a swarm is, how a role run is designed to work, and what you can use today.

<Info>
  **Upcoming.** This page describes the full agent swarm: a role run that
  carries every candidate from first contact to a decision-ready report, and
  raises reports to your [Agent Inbox](/smaya/agent-inbox). Starting a role run,
  the Inbox items it raises and every step of the run are Upcoming. What you can
  use today is screen-to-rank (EX Call, EX Voice and EX Avatar through scoring
  to the Leaderboard), [AI Employees](/smaya/ai-employees) in chat, Smaya ask
  and report chat, and the [MCP server](/smaya/mcp). See [What is Available
  today](#what-is-available-today).
</Info>

## Purpose

An **agent swarm** is how Smaya will run hiring for a role. You give it the role. It builds the pipeline, talks to every candidate, gathers the evidence, and brings you the people worth deciding on.

You see outcomes and reports. You do not manage agents. The work between reports runs in the background, and you step in when there is something to decide.

## What a swarm is

A swarm is **one run per role, plus one sub-run per candidate**. Each run is built from pre-built agents in the [Agent Library](/smaya/capabilities). Each sub-run carries one candidate from first contact to a decision-ready report.

<CardGroup cols={3}>
  <Card title="One instruction" icon="play">
    You start a role with the role, your notes and, if you have them, example profiles. That is the last thing you are asked until a report exists.
  </Card>

  <Card title="Many candidates at once" icon="users">
    Every candidate moves through the same steps at the same time, on the same
    criteria. Candidate one and candidate fifty stay comparable.
  </Card>

  <Card title="Reports, not tasks" icon="file-check">
    You will never approve an email or a slot request. You answer reports and accept schedules, and nothing else.
  </Card>
</CardGroup>

## How a role runs, end to end

```mermaid theme={null}
%%{init: {'theme':'base','themeVariables':{'primaryColor':'#EEF2FF','primaryBorderColor':'#4B6CFE','primaryTextColor':'#0F172A','lineColor':'#64748B','clusterBkg':'#F8FAFC','clusterBorder':'#E2E8F0'}}}%%
flowchart TD
  ROLE([A published job with locked scoring])
  ROLE --> CALL([EX Call phone screen])
  CALL --> AI([EX Voice or EX Avatar interview])
  AI --> SCORE([Scored against the locked criteria])
  SCORE --> RANK([Ranked on the Leaderboard])
  RANK --> YOU([A person decides, with a reason])
  RUN(["Run the whole role for you (Upcoming)"]) -.-> SRC(["Profile identified (Upcoming)"])
  SRC -.-> OUT(["Outreach reply (Upcoming)"])
  OUT -.-> CALL
  AI -.-> INBOX(["Reports in your Agent Inbox (Upcoming)"])
  RANK -.-> PANEL(["Panel availability (Upcoming)"])
  INBOX -.-> YOU
```

*Solid lines work today. Dotted lines are Upcoming parts of the full role run.*

<Steps>
  <Step title="Run the role (Upcoming)">
    You give the role in one instruction, from Exterview or from a connected assistant. The scoring criteria are locked for the run, so every candidate is assessed the same way. Today, criteria are already locked per job when you create it.
  </Step>

  <Step title="Candidates move through the pipeline">
    Today, each candidate is screened with EX Call, EX Voice or EX Avatar, scored
    against the locked criteria and ranked on the Leaderboard. Upcoming: finding,
    researching and reaching out to candidates as steps of the run.
  </Step>

  <Step title="Reports reach your Agent Inbox (Upcoming)">
    Upcoming: at five moments in each candidate's journey, the swarm raises a
    report. When it needs a person's time, it raises a schedule.
  </Step>

  <Step title="You answer (Upcoming)">
    Proceed, Drop or Talk on a report. Accept, Decline or Talk on a schedule. Each answer is recorded with who gave it and why. Today, you review reports in the Decision Center and change a decision with a stated reason.
  </Step>
</Steps>

## The five report moments

As designed, every candidate can produce up to five reports. Each one tells you what was found or concluded, with the evidence behind it. All five are Upcoming as Inbox items. The underlying phone screen and AI interview reports are Available today in the Decision Center.

| Moment                        | What you will see                                               | If you do not answer (Upcoming)  |
| ----------------------------- | --------------------------------------------------------------- | -------------------------------- |
| **1. Profile identified**     | Who the candidate is, why they fit, and the fit score           | The candidate continues          |
| **2. Outreach reply**         | What the candidate said, their interest, and consent to call    | The candidate continues          |
| **3. Phone call taken**       | The call summary and transcript                                 | The candidate continues          |
| **4. AI interview conducted** | The scored report: dimensions, evidence, criteria version, rank | The candidate waits for you      |
| **5. Panel availability**     | A proposed panel slot the candidate has confirmed               | The slot waits for you to accept |

As designed, the first three continue by default so candidates are not left waiting. The AI interview report and every schedule wait for a person, because that is where the decision and your team's time sit. A Drop stops a candidate at any moment. See [Agent Inbox](/smaya/agent-inbox).

## What runs on its own

As designed, the swarm works between reports without asking you anything.

<CardGroup cols={2}>
  <Card title="Finding and researching" icon="search">
    **Upcoming.** Sourcing candidates for the role and building a picture of each one: experience, fit and the best way to reach them.
  </Card>

  <Card title="Reaching out" icon="send">
    **Upcoming.** Contacting candidates on the channel their region allows,
    following up, and reading their replies.
  </Card>

  <Card title="Screening and interviewing" icon="phone">
    **Available.** Phone screens with EX Call, AI interviews with EX Voice or EX
    Avatar, each producing a report.
  </Card>

  <Card title="Scoring and ranking" icon="chart-line">
    **Available.** Scoring against the locked criteria and keeping the Leaderboard current as interviews complete.
  </Card>
</CardGroup>

Today you trigger each screen and interview yourself, or ask an AI Employee to. Upcoming: the run doing it for every candidate on its own.

Setup decisions — the scoring criteria, the interview design, the agent's role for the job — happen before the run starts. They are never Inbox items.

## Human oversight that does not move

Autonomy between reports will never move a hiring decision to a machine. These rules hold today and apply to every run.

<CardGroup cols={2}>
  <Card title="People make hiring decisions" icon="scale">
    No automated step sets a candidate to hired, rejected or withdrawn. The swarm recommends. A person decides.
  </Card>

  <Card title="Drop is a human decision on record" icon="user-check">
    A Drop moves the candidate out of the queue with who decided and why. It never
    marks the candidate rejected.
  </Card>

  <Card title="Notice before every AI interview" icon="megaphone">
    A candidate confirms they have been told AI is part of the interview before
    every AI interview starts. No automated step can skip it.
  </Card>

  <Card title="Nobody is hidden" icon="eye">
    Candidates below the bar stay visible for a person to review. Nothing suppresses a result.
  </Card>
</CardGroup>

As designed, every decision that moves a candidate in a run is recorded with who made it: a person, a time-window policy, or the candidate. See [Governance & Control](/smaya/oversight).

## Channel choice follows the candidate's region

**Upcoming.** Which channel the swarm may use first will depend on **where the candidate is**, not where your organization is. A candidate whose region is not known gets the strictest rule.

| Candidate region     | First contact                                            | AI phone call                                  |
| -------------------- | -------------------------------------------------------- | ---------------------------------------------- |
| **India**            | Email, WhatsApp, or an AI call after a do-not-call check | Allowed as first contact after that check      |
| **United States**    | Email, or SMS where the candidate opted in               | Only after the candidate agrees in their reply |
| **United Kingdom**   | Email                                                    | Only after the candidate agrees in their reply |
| **European Union**   | Email                                                    | Only after the candidate agrees in their reply |
| **Other or unknown** | Email                                                    | Not until the region is known                  |

Every channel decision will be recorded with the region, the rule version and any consent the candidate gave. This is a product rule, not legal advice. Your counsel decides what applies to your organization, and Exterview makes no claim of compliance with any regulation.

## What is Available today

| Capability                                                                            | Status        |
| ------------------------------------------------------------------------------------- | ------------- |
| Screen to rank: EX Call, EX Voice and EX Avatar through scoring to the Leaderboard    | **Available** |
| [AI Employees](/smaya/ai-employees) in chat, with handoffs and approval before writes | **Available** |
| Smaya ask and report chat                                                             | **Available** |
| Ask from Claude, ChatGPT or Microsoft 365 Copilot over [MCP](/smaya/mcp)              | **Available** |
| Criteria locked per job so candidates stay comparable                                 | **Available** |
| Score overrides with a stated reason, and a record of the model behind each score     | **Available** |
| Memory across your conversations with Smaya                                           | **Available** |
| Start a role run, from Exterview or a connected assistant                             | **Upcoming**  |
| [Agent Inbox](/smaya/agent-inbox) items: reports to view and schedules to accept      | **Upcoming**  |
| The full role-to-decision run across all five report moments                          | **Upcoming**  |
| Ask an AI Employee from Slack                                                         | **Upcoming**  |
| Agent Catalog and Agent Lineage                                                       | **Upcoming**  |
| Sourcing and a research packet for every candidate                                    | **Upcoming**  |
| Outreach sequences (Outreach Campaigns)                                               | **Upcoming**  |
| Region-based channel choice                                                           | **Upcoming**  |
| Screening that starts the moment a candidate applies                                  | **Upcoming**  |
| Continue-by-policy time windows at every report moment                                | **Upcoming**  |
| Talk: booking a call with the candidate from an Inbox answer                          | **Upcoming**  |
| Swarm Control Plane: live view of every run                                           | **Upcoming**  |
| Round-the-clock operation on a standing goal                                          | **Upcoming**  |
| Kill switch and pause per run                                                         | **Upcoming**  |
| Per-organization budget limits                                                        | **Upcoming**  |
| Interviewer simulator                                                                 | **Upcoming**  |
| Feedback loop from overrides to new criteria versions                                 | **Upcoming**  |
| Trace export to your own tools                                                        | **Upcoming**  |
| Agent Inbox cards in Microsoft Teams                                                  | **Upcoming**  |

## FAQs

<AccordionGroup>
  <Accordion title="Can I run a whole role with the swarm today?">
    Not yet. The role run is Upcoming. Today you screen, interview, score and rank candidates yourself or with an AI Employee, and decide in the Decision Center.
  </Accordion>

  <Accordion title="Do I have to manage the agents in a swarm?">
    No. You will give the role and answer reports. Which agents do the work is the
    platform's job.
  </Accordion>

  <Accordion title="Can the swarm reject a candidate?">
    No. Only a person can set a hiring outcome. A Drop in the Agent Inbox is your
    decision, recorded with your reason, and it never marks the candidate
    rejected.
  </Accordion>

  <Accordion title="Will the swarm contact candidates without asking me?">
    Not today: outreach is Upcoming. As designed, yes, on the channel the
    candidate's region allows. You set the criteria and the agent's role for the
    job before the run starts.
  </Accordion>

  <Accordion title="Can criteria change during a run?">
    No. Criteria are locked for the run. A change applies to the next run.
  </Accordion>
</AccordionGroup>

## Related

<div className="kb-links">
  *   <span className="kb-ico">
        <Icon icon="inbox" />
      </span>
    [Agent Inbox](/smaya/agent-inbox). Where you will answer reports and
    schedules.
  *   <span className="kb-ico">
        <Icon icon="id-card" />
      </span>
    [AI Employees](/smaya/ai-employees). The roles that do the work.
  *   <span className="kb-ico">
        <Icon icon="shield-check" />
      </span>
    [Governance & Control](/smaya/oversight). The controls every run inherits.
</div>