> ## 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.

# Finance

> The agent that watches the cash — captures expenses, reads the numbers, flags what looks off. And never moves money without you.

It's 11pm on a Tuesday. The month closed with a gap nobody can explain.
You open the spreadsheet and find three uncategorized entries, an invoice
someone forgot to log, and a subscription that doubled without warning.
There's a partners' meeting tomorrow and the question will be "how are we
tracking this month?" — and you still can't answer it.

What leaves your plate: the **work of collecting, organizing, and reading**
the numbers. An agent watches the cash all day, turns a receipt into an
entry, reads the trend, and tells you when something looks off. **What it
never does is move money** — pay, refund, approve. That stays yours, one
tap away.

<Note>
  **The rule that organizes this whole page:** the agent **reads and
  prepares**; it **does not pay**. Anything that touches money becomes a
  proposal that waits for your yes. The tedious part is its job; the
  decision is yours.
</Note>

## The pattern, applied to cash

Every flow here follows the same shape as the rest of Apollo — only with
the handbrake always on when money is involved:

```mermaid theme={null}
flowchart LR
    A[Trigger<br/>receipt, deadline, question] --> B[Agent pulls context<br/>from the Company Brain]
    B --> C[Reads the numbers + uses tools<br/>Finance, docs, integrations]
    C --> D{Touches money?}
    D -->|No: read, organize, summarize| E[Does it and logs it]
    D -->|Yes: pay, refund, approve| F[Proposes + waits for you]
    E --> G[Becomes memory<br/>next month is cleaner]
    F --> G
```

## Four everyday flows

### 1. A forwarded receipt becomes an entry

You forward the agent a photo of the lunch receipt from the client
meeting, or a PDF invoice that landed in your inbox.

* **Trigger:** a receipt / message with an amount reaches the agent.
* **Context:** it reads the document and checks the [Company Brain](/en/features/brain)
  to recognize the vendor ("this is the same SaaS we pay every month")
  and the category you usually use.
* **Tools:** it logs an entry in [Finance](/en/features/financeiro) with
  amount, date, vendor, and category pre-filled; attaches the receipt as
  proof.
* **Waits for you:** the entry lands as a **draft for review**. You check
  the category with one tap and confirm — the agent never calls the books
  closed on its own.
* **Becomes memory:** next time that vendor shows up, the category is
  already right.

### 2. The routine that summarizes spend and flags anomalies

A [routine](/en/features/rotinas) the agent runs on its own, every Monday
morning.

* **Trigger:** the scheduled time of the weekly routine.
* **Context:** it reads the week's entries and compares them with prior
  weeks stored in the Brain — what's normal, what's off the curve.
* **Tools:** it builds a short summary ("this week's spend: X on
  operations, Y on marketing; one subscription is up 40% vs. its
  history") and raises a [proactive notification](/en/features/chats) in
  your inbox.
* **Waits for you:** when it sees a value that stands out, it **flags it
  for review** — it doesn't cancel the subscription or dispute the
  charge. It points and asks.
* **Becomes memory:** what you mark as "expected" stops being an alarm;
  the agent calibrates noise versus what deserves your attention.

### 3. "How are we tracking this month?" — answered on the spot

The everyday question, one tap from the [Chief of Staff](/en/agents/athena).

* **Trigger:** you ask in chat "how's the cash this month?".
* **Context:** the agent reads the period's entries straight from Finance
  and the prior-month history in the Brain.
* **Tools:** it returns revenue, spend, and the trend vs. last month,
  with the biggest inflows and outflows named — without you opening three
  tabs.
* **Waits for you:** nothing to approve here — it's pure reading. But
  every answer carries the trail of where each number came from, so you
  can audit it if you want.
* **Becomes memory:** the questions you ask most teach the agent to have
  the cut that matters to you ready in advance.

### 4. A reconciliation list, ready for human review

Month-end without the manual digging.

* **Trigger:** the close-out routine, or you asking "prepare this
  month's reconciliation".
* **Context:** the agent cross-checks entries against attached receipts
  and marks what doesn't line up — no receipt, no category, duplicate
  amount, unknown vendor.
* **Tools:** it builds a [document](/en/features/documentos) or a list of
  [tasks](/en/features/boards) with each open item and a suggested owner.
* **Waits for you:** the list is a **review proposal**. Nothing is
  reconciled, written off, or paid — the human decides item by item.
* **Becomes memory:** the patterns you correct (this vendor is always
  "operations") shorten next month's list.

### Bonus: remind owners of pending entries

The agent also chases, politely. When someone left an advance without
clearing it or a receipt unposted, it raises a reminder in the inbox and,
if you configure it, sends a nudge over [WhatsApp](/en/integrations/whatsapp).
It chases the *record* — never the payment.

## What stays human

<Warning>
  **Everything that moves money is proposed and waits for a human. The
  agent reads and prepares; it does not pay.** These always stay in your
  hands:

  * **Paying, transferring, or refunding** any amount.
  * **Approving an expense** or releasing an advance.
  * **Closing the books** or marking an entry reconciled.
  * **Disputing a charge** or canceling a vendor contract.
  * **External comms involving amounts** (billing a client, negotiating
    with a vendor).
  * **Sharing financial numbers** beyond leadership.

  The agent hands everything ready — the checked entry, the reconciled
  list, the summary with the off-curve number in the spotlight. The click
  that moves the money is yours, recorded in an audit trail.
</Warning>

<Info>
  **Why the brake is firmer here.** In other areas the agent acts and
  only the sensitive part waits. In finance the rule is reversed: the
  default is to **propose**, and only reading/organizing does it resolve
  on its own. Money moved wrong has no "ctrl-z" — so autonomy on cash
  actions is earned per action-class, never a blank check.
</Info>

## Where to start

<Steps>
  <Step title="Give the agent the history">
    Upload the last few months of expenses and your recurring receipts to
    the [Company Brain](/en/features/brain). That's what teaches the agent
    to recognize vendor and category — without it, every entry becomes a
    question.
  </Step>

  <Step title="Forward a receipt and check the draft">
    Send the agent a receipt and watch the entry come back pre-filled.
    Check the category, confirm. Do it three or four times — that's how it
    calibrates to the way you categorize.
  </Step>

  <Step title="Schedule the weekly routine and set the caps">
    Ask for a summary + anomalies routine every Monday, and set
    [Stars caps](/en/billing/stars) per org and per agent. The budget
    stops spend before the call — Apollo's own cash handled with the same
    care as yours.
  </Step>
</Steps>

## Next steps

<CardGroup cols={2}>
  <Card title="Finance (feature)" icon="wallet" href="/en/features/financeiro">
    The records themselves — entries, expenses, cashflow, and who sees what.
  </Card>

  <Card title="Company Brain" icon="brain" href="/en/features/brain">
    The memory that lets the agent recognize vendor and category.
  </Card>

  <Card title="Athena — Chief of Staff" icon="user-tie" href="/en/agents/athena">
    The operator that answers "how are we tracking this month?" on the spot.
  </Card>

  <Card title="Backoffice" icon="folder-open" href="/en/use-cases/backoffice">
    Cash inside the bigger picture of administrative routine.
  </Card>
</CardGroup>
