# Receiptor Agents

Receiptor gives agents a bookkeeping operating layer, not just document OCR or a thin tool wrapper.

It combines:

- Context: workspaces, extracted documents, integrations, saved views, export history, and prior categorization decisions
- Automations: workflow automations, routing logic, review steps, and account or workspace permissions
- Agents: a CLI and MCP-friendly capability layer for receipt intake, document search, categorization review, workspace setup checks, and spend analysis

## Operating model

Financial inputs become structured bookkeeping context. Receiptor keeps the historical record, learned memory, and bookkeeping state that agents need, then applies automations, skills, and permissions so work can move forward without reconstructing business logic from scratch.

```text
                                  RECEIPTOR

        +---------------------------- Context -----------------------------+
        |                                                                  |
        |  Memories                                                        |
        |  Historical context                                              |
        |  Past decisions                                                  |
        |  Corrections                                                     |
        |  Workspace state                                                 |
        |                                                                  |
        |             +----------------- Automations ----------------+      |
        |                |                                       |          |
        |                |  Automations                          |          |
        |                |  Skills                               |          |
        |                |  Permissions (soon)                   |          |
        |                |                                       |          |
        |                |        +-------- Agents --------+     |          |
        |                |        |                        |     |          |
        |                |        +------------------------+     |          |
        |                |                                       |          |
        |                +---------------------------------------+          |
        |                                                                  |
        +------------------------------------------------------------------+

customer docs                                                       books
inboxes              <----->                            <----->     exports
feedback                                                          downstream
```

## Context

- Receipts, invoices, merchants, amounts, and categories
- Memories and historical bookkeeping context
- Workspace-scoped bookkeeping state
- Connected sources, integrations, destinations, and exports

## Automations

- Event and schedule-based automations
- Reusable skills for recurring finance workflows
- Routing and destination logic
- Workspace permissions and secure account boundaries, with more permission controls coming

Agents should work inside those controls instead of reconstructing finance logic from a blank prompt each time.

## Install

```bash
curl -fsSL https://receiptor.ai/install.sh | bash
```

## Authenticate

```bash
receiptor auth sk_your_api_key
receiptor workspace list
receiptor workspace use org_123
```

## Core commands

```bash
receiptor list
receiptor documents search --query "uber" --limit 5
receiptor files upload ./receipt.pdf
receiptor workspaces list
```

## Built-in skills

- `setup-readiness`: check auth, workspaces, and integration readiness before an agent starts operating
- `receipt-intake`: upload and inspect receipts and invoices
- `categorization-review`: review and correct classification or coding issues
- `spend-analysis`: pull grouped totals and reporting-ready spend outputs

## Example flow

- A new receipt enters a workspace from email or upload
- Receiptor applies the right automation, routing logic, and permissions
- An agent can inspect the document, review categorization, find related spend, and move the workflow forward using structured tools

## Notes

- Most meaningful operations require a Receiptor AI account and workspace.
- Uploads and some extraction workflows are asynchronous.
- The CLI is a thin client over the Receiptor capability API, so available commands depend on the authenticated account, workspace context, and automations that govern that workspace.

## Links

- Install script: `https://receiptor.ai/install.sh`
- Docs: `https://docs.receiptor.ai`
- Website: `https://receiptor.ai`