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

# Agents

> What an agent is in Apollo Space: persona, tools, memory, traces, and budget.

## Anatomy of an agent

An agent in Apollo Space is the sum of four things:

<CardGroup cols={2}>
  <Card title="Persona" icon="masks-theater">
    Tone, voice, and behavior. Defines how the agent responds, in what
    language and with what level of formality — and what it knows about
    its organization.
  </Card>

  <Card title="Tools" icon="screwdriver-wrench">
    What the agent can do — search the web, send WhatsApp messages,
    create a lead, read Notion. Each tool is a typed function with its
    own permissions.
  </Card>

  <Card title="Memory" icon="brain">
    Knowledge that persists across conversations. Can be **factual**
    (dates, preferences), **procedural** (approaches that worked),
    or **semantic** (insights distilled from many conversations).
  </Card>

  <Card title="Budget" icon="star">
    Spend caps configurable by the org — a safeguard so an unexpected
    loop doesn't drain the entire balance. See
    [Caps](/en/billing/caps).
  </Card>
</CardGroup>

## What happens in a conversation

When you send the agent a message, Apollo Space runs the following
sequence under the hood:

```mermaid theme={null}
flowchart TD
    A[User message] --> B[Opens trace]
    B --> C{Within budget?}
    C -->|Blocked| Z[Response: cap reached]
    C -->|OK| D[Agent selects tools]
    D --> E{Tools needed?}
    E -->|Yes| F[Runs tools with org permissions]
    E -->|No| G[Composes text directly]
    F --> H[Records usage + debits Stars]
    G --> H
    H --> I[Replies in chat + closes trace]
```

<Steps>
  <Step title="Opens the trace">
    A **trace** begins — the complete record of everything the agent did
    in that conversation. Every action, cost, and return value is logged
    for replay and auditing.
  </Step>

  <Step title="Checks the budget">
    Before any billable action, Apollo Space checks whether the call fits
    within the limits — wallet balance plus caps by org, agent, and
    user. See [Caps](/en/billing/caps).
  </Step>

  <Step title="Agent selects tools">
    The agent decides which tools to call (or none, if the response is
    plain text). Apollo Space runs each one with the agent's permissions
    and your org's isolation.
  </Step>

  <Step title="Records usage">
    Each billable action becomes a line in the org's usage history.
    Stars are debited from the wallet in the same operation — no
    separate accounting.
  </Step>

  <Step title="Closes the trace">
    The response returns to the chat. The trace is available via
    **View trace** — you can inspect it anytime afterward.
  </Step>
</Steps>

## The agent hierarchy

Agents aren't a flat list — they form an **org chart**, just like a
real company. Each level delegates downward and reports upward:

```mermaid theme={null}
flowchart TD
    DT["Digital Twin<br/>(personal — spans all your orgs)"] --> CoS["Chief of Staff<br/>(the organization's agent)"]
    CoS --> M["Functional managers<br/>(sales, finance, operations...)"]
    M --> S["Specialists<br/>(research, outbound, writing, analysis...)"]
```

* **[Digital Twin](/en/agents/digital-twins)** — your **personal** agent,
  spanning every organization you belong to and speaking with **your**
  voice. It's the headline feature of Apollo.
* **[Chief of Staff](/en/agents/athena)** (Athena) — the **organization's**
  agent: coordinates company work with a neutral voice.
* **Managers and specialists** — function- and task-specific personas
  that the Chief of Staff activates as needed.

<Info>
  The Digital Twin and the Chief of Staff are **distinct** agents with
  different roles — one represents the person, the other runs the
  company. They collaborate, but they are not "the same agent in two
  modes".
</Info>

## Agent types

Apollo Space ships with **ready-made agents** and also lets you create
**custom agents**.

### Canonical personas

Apollo Space ships with personas that cover day-to-day work — from
research to closing, from operations to back-office. The three most
used:

<Tabs>
  <Tab title="Scout — Researcher">
    <CardGroup cols={2}>
      <Card title="What it does" icon="binoculars">
        Public web search via Tavily, company analysis, information
        consolidation with cited sources. Ideal for the research and
        lead enrichment phase.
      </Card>

      <Card title="Main tools" icon="screwdriver-wrench">
        Web search (Tavily), page reading, contact extraction via
        Apify, public profile analysis.
      </Card>
    </CardGroup>

    <Info>
      Every Scout search includes links to the sources consulted —
      you can verify the origin of each piece of information.
    </Info>

    [See Scout in detail →](/en/agents/scout)
  </Tab>

  <Tab title="Marcus — Outbound">
    <CardGroup cols={2}>
      <Card title="What it does" icon="paper-plane">
        Writes and sends email + WhatsApp sequences, landing in the
        recipient's real inbox. Optional human approval before each
        send.
      </Card>

      <Card title="Main tools" icon="screwdriver-wrench">
        Email sending (Resend), WhatsApp via Twilio BYO-key,
        sequence management, CRM activity logging.
      </Card>
    </CardGroup>

    <Warning>
      To send emails or WhatsApp messages, Marcus needs at least one
      channel connected. See [Set up WhatsApp](/en/guides/set-up-whatsapp).
    </Warning>

    [See Marcus in detail →](/en/agents/marcus)
  </Tab>

  <Tab title="Athena — Chief of Staff">
    <CardGroup cols={2}>
      <Card title="What it does" icon="user-tie">
        The organization's multi-tool operator — stitching together
        tasks across the CRM, external integrations, and daily
        actions, with the company's neutral voice.
      </Card>

      <Card title="Main tools" icon="screwdriver-wrench">
        Full CRM access (leads, pipelines, activities), hundreds of
        integrations via Composio, task creation, context summaries.
      </Card>
    </CardGroup>

    <Info>
      Athena is the **organization's Chief of Staff** — neutral voice,
      runs the company. For the **personal** agent that speaks with a
      leader's voice, see the [Digital Twin](/en/agents/digital-twins)
      (a distinct agent).
    </Info>

    [See Athena in detail →](/en/agents/athena) · [Digital Twins →](/en/agents/digital-twins)
  </Tab>
</Tabs>

### Chief of Staff vs. Digital Twin

These are **two distinct agents** with different roles — not "two modes
of the same agent":

<CardGroup cols={2}>
  <Card title="Chief of Staff (Athena)" icon="building" href="/en/agents/athena">
    The **organization's** agent. Knows the company's context and
    coordinates its work, but speaks with a neutral voice — not "as"
    any specific leader.
  </Card>

  <Card title="Digital Twin" icon="user-astronaut" href="/en/agents/digital-twins">
    The **person's** agent — personal, spans all their orgs, and speaks
    with the leader's **voice**. Shows up in conversations they don't
    have time to handle.
  </Card>
</CardGroup>

<Info>
  They **collaborate**: a leader's Twin delegates to Athena when the
  work belongs to the company; Athena activates the Twin when a
  communication needs to go out in that person's voice. Distinct roles,
  one team.
</Info>

### Other available personas

Beyond the three canonical personas above, Apollo Space offers a
**catalog of specialist personas** that are opt-in per org. It includes
function-specific Chiefs of Staff (finance, technology, marketing) and
agents focused on narrower tasks — writing, design, post-sale
relationships, competitive analysis, and more.

These personas are **opt-in** — you enable only the ones that make
sense for your team. There's no cost to "keep an agent alive" — cost is
always per actual call made. The full list is visible under
**Settings → Agents → Catalog** inside the app.

### Custom agents

Each org can create its own agents with:

* **Custom persona**: your own system prompt plus defined tone and voice
* **Custom tools**: you enable only the tools this agent needs
* **Custom caps**: can have a daily cap different from the org's default
* **Own memory**: agents don't share memory — each accumulates its own
  history

See [Create a custom agent](/en/guides/invite-team) (coming soon) for the
step-by-step guide.

## Audit trail

Every agent action generates two types of log:

| Log               | Purpose                                                                                                            |
| ----------------- | ------------------------------------------------------------------------------------------------------------------ |
| **Trace**         | Complete replay of the conversation — every decision, every tool call. Visible in the UI via "View trace" in chat. |
| **Usage history** | One line per billable action, with cost. Feeds the Billing panel and the caps.                                     |

The audit log is **append-only** — nothing is deleted when a
conversation ends.

## Next steps

<CardGroup cols={2}>
  <Card title="Scout" icon="binoculars" href="/en/agents/scout">
    Details on the research agent.
  </Card>

  <Card title="Stars + caps" icon="star" href="/en/billing/stars">
    How to size the budget per agent.
  </Card>

  <Card title="Integrations" icon="puzzle-piece" href="/en/integrations/composio">
    Which external tools your agents can use.
  </Card>

  <Card title="CRM" icon="users" href="/en/features/crm">
    How agents interact with leads and pipelines.
  </Card>
</CardGroup>