Follow the invoice inbox

# Invoice inbox

## Intake or invoice: the difference

An **intake** is a received supplier document (invoice, credit note or receipt) while it is under review in the invoice inbox: it is the step **before** accounting. Once processed, it becomes an invoice (`get_invoices`) and then a payable (`get_payables`).

- "Did we receive the invoice from X?" → `get_intakes`
- "Is it approved, paid?" → `get_invoices` (see the invoices waiting for approval skill)

## Step 1 — What is waiting in the inbox

```json
{ "tool": "get_intakes", "params": { "stage": ["inInbox"], "pageSize": 100 } }
```

`stage` is the simple view of the lifecycle:

| `stage` | Meaning |
|---------|---------|
| `inInbox` | Awaiting action or under review |
| `linked` | Attached to an existing document |
| `processed` | Finished (validated, refused, deleted) |

⚠️ Traps verified live:
- **An empty list only proves "nothing in this stage"**. On the demo account, `inInbox` returned 0 documents while a processed document existed: it only appeared when `processed` was requested.
- Archived intakes are excluded unless you ask for `stage: ["processed"]` or `status: ["deleted"]`. A document missing from the list does not prove it does not exist.
- There is **no type filter**: invoices, credit notes and receipts are mixed. Read `type` on each row before aggregating.
- Fields extracted from the document (supplier, amounts, dates, number) are `null` until captured — `null` is not zero.
- Amounts are in the document's own currency, with no conversion: group by currency, never add them together.
- The `createdAfter`/`createdBefore` filters expect a full ISO datetime (`2026-10-01T00:00:00.000Z`) and apply to when Spendesk received the document, not to the invoice date.

## Step 2 — Who it is assigned to

`assigneeId` is a raw ID. Resolve it with:

```json
{ "tool": "get_users", "params": { "ids": ["<assigneeId>"] } }
```

Never show the bare ID; if it does not resolve, say so.

## Step 3 — One document in detail, and what it became

```json
{ "tool": "get_intake_by_id", "params": { "id": "<intake id>" } }
```

On top of the list fields, it returns:
- `lineItems` and `lineItemsSource` (e.g. `ocr`): what was actually billed;
- `documents`: the received files, with a short-lived download link;
- `payableId` and `resultingDocument`: what the document produced once processed. `payableId` resolves with `get_payable_by_id`.

`get_intake_by_id` also finds archived intakes that the list leaves out. `appUrl` is the direct link to the document in Spendesk — give it as returned, never build it.

## Recommended output format

```
Invoice inbox — <today's date>
Waiting: N documents (per currency: …)

| Received on | Type | Supplier | No. | Amount | Due date | Assigned to | Status |
```

Sort by due date, earliest first, and flag documents due within 7 days that are not processed yet.

## Observed result (demo account)

Inbox empty under `inInbox`. One document under `processed`: an invoice received by email on 15/09/2026, status `validated`, EUR 15,779.40 gross (EUR 13,260.00 net, 19% VAT read by OCR), due 06/10/2026, assigned to a user resolved with `get_users`, with a `payableId` set.

## Example user prompts

- "What is waiting in the invoice inbox?"
- "Did we receive invoice RE-2026-08842, and where does it stand?"
- "Which documents received this week are not processed yet, and who are they assigned to?"