---
title: "LLM enrichment"
description: "Optional add-on: run an extra model pass with your own prompt over text, tables, or figures."
---

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

# LLM enrichment

> **Warning:** **We do not recommend LLM enrichment for most workloads.** r-1 output is accurate by default. An extra model pass adds latency, costs more, and can make the output less accurate. Try plain r-1 first, and use enrichment only when you need the output in a specific form that r-1 does not give you.

LLM enrichment is an optional add-on. It runs an extra model pass with your own `prompt` over form regions, tables, or figures, and changes how that content is written in the output. Pages without a matching block continue through r-1 unchanged. See [Credit usage](/reference/credit-usage) for the price of prompted blocks.

## Add a prompt

Add an entry to `enhance.agentic` with a `scope` and a `prompt`:

**Python**

```python
result = client.parse.run(
    input=upload.file_id,
    settings={"model": "r-1"},
    enhance={
        "agentic": [{
            "scope": "table",
            "prompt": "Write each footnote marker as [n] after the value it belongs to.",
        }]
    },
)
```

**Node.js**

```javascript
const result = await client.parse.run({
  input: upload.file_id,
  settings: { model: 'r-1' },
  enhance: {
    agentic: [{
      scope: 'table',
      prompt: 'Write each footnote marker as [n] after the value it belongs to.',
    }],
  },
});
```

**cURL**

```bash
curl -X POST https://platform.reducto.ai/parse \
  -H "Authorization: Bearer $REDUCTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "input": "reducto://your-file-id",
    "settings": {"model": "r-1"},
    "enhance": {
      "agentic": [{
        "scope": "table",
        "prompt": "Write each footnote marker as [n] after the value it belongs to."
      }]
    }
  }'
```

## Scopes

| `scope` | Applies to | Options |
| --- | --- | --- |
| `"text"` | Form regions, such as key-value fields. Not body text. | `prompt` |
| `"table"` | Tables. | `prompt`, `mode`: `"default"` runs on tables likely to benefit, `"max"` runs on every table. |
| `"figure"` | Figures. | `prompt`, `advanced_chart_agent` (see [Advanced chart extraction](/configs/parse/chart-extraction)). |

You can add one entry per scope in the same request.

An `enhance.agentic` entry without a `prompt` is a legacy Parse feature and does nothing useful on r-1. See [Legacy Parse settings](/configs/parse/legacy).

Source: https://docs.reducto.ai/configs/parse/enrichment/index.mdx
