# Pulse AI

> Read documents with Pulse AI: turn PDFs, scans, photos, spreadsheets and Office files into text, pull out named fields and tables, split long documents by topic, sort documents by kind, and fill or clear PDF forms.

App · 14 actions. Web page: https://docs.lodol.com/docs/integrations/pulse-ai

## Actions

### Cancel job

Stop a Pulse job that is still waiting or running. Pages read before it stops still count. A job that has already ended is left as it is.

**Inputs**

- Job ID (`job_id`, text, required): The job_id from 'Start document extraction' or any other step.

**Outputs**

- `canceled` (true/false): False when the job had already ended
- `job_id` (text): The job's id
- `message` (text): Pulse's answer

**Example output**

```json
{
  "canceled": true,
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "message": "Job cancelled successfully"
}
```

### Classify document

Tell which of the kinds of document you list a file is, from its first pages, for example to send invoices and contracts down different paths. Always picks one, so include a catch-all. Uses 0.5 credits a page.

**Inputs**

- File (`file`, text, required): The document: upload it, pick a file from an earlier step, or give a link Pulse can download. PDFs, images (JPG, PNG, TIFF and more), Word, PowerPoint, Excel, CSV and HTML files work. Uploads can be up to 100 MB.
- Kinds of document (`categories`, object, required): Each kind's name with what it looks like, for example {"invoice": "A bill asking for payment", "other": "Anything that is not an invoice"}.
- Pages (`pages`, text): The pages to look at, such as 1-3. Pulse's default is the first 5.

**Outputs**

- `classification` (text): The kind of document it is: one of your names
- `credits_used` (number): Credits this step used (empty when your plan doesn't use credits)
- `job_id` (text): The Pulse job that did the work
- `page_range` (text): The pages that were looked at

**Example output**

```json
{
  "classification": "invoice",
  "credits_used": 1.0,
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "page_range": "1-2"
}
```

### Clear form

Erase what was written or ticked on a PDF form, keeping the printed blank form, and save it. Must be turned on for your organization by Pulse. Uses 3 credits a page.

**Inputs**

- PDF form (`file`, text, required): The PDF form: upload it, pick a file from an earlier step, or give a link Pulse can download.
- What to clear (`instructions`, text): Only clear these fields. Leave empty to clear them all.
- Pages (`pages`, text): Only these pages, such as 1-3,5. Leave empty for the whole document.

**Outputs**

- `content_type` (text): The saved file's type (application/pdf)
- `credits_used` (number): Credits this step used (empty when your plan doesn't use credits)
- `document_ref` (text): The saved PDF, for a later step to attach
- `fields_cleared` (number): How many fields were emptied
- `filename` (text): The saved PDF's name
- `form_id` (text): The cleared form's id in Pulse
- `job_id` (text): The Pulse job that did the work
- `page_count` (number): How many pages the PDF has
- `size_bytes` (number): The saved PDF's size in bytes
- `url` (text): Where to download the saved PDF in Lodol

**Example output**

```json
{
  "content_type": "application/pdf",
  "credits_used": 3.0,
  "document_ref": "doc_0123456789abcdef0123456789abcdef",
  "fields_cleared": 7,
  "filename": "pulse-form-f1b2a3d4.pdf",
  "form_id": "f1b2a3d4-0000-4000-8000-00000000000a",
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "page_count": 1,
  "size_bytes": 140815,
  "url": "/actions/core/serve-document/doc_0123456789abcdef0123456789abcdef"
}
```

### Extract document

Read a document (PDF, scan, photo, spreadsheet, Word or PowerPoint file) and return its text as Markdown, tables included. Waits until Pulse has finished. The extraction_id it returns feeds the other steps. Uses 1 credit a page.

**Inputs**

- File (`file`, text, required): The document: upload it, pick a file from an earlier step, or give a link Pulse can download. PDFs, images (JPG, PNG, TIFF and more), Word, PowerPoint, Excel, CSV and HTML files work. Uploads can be up to 100 MB.
- Pages (`pages`, text): Only these pages, such as 1-3,5. Leave empty for the whole document.
- About the document (`instructions`, text): Anything that helps Pulse read it, such as what kind of document it is or how to treat its numbers.
- Double-check (`refine`, list): Extra passes that correct what was read, for hard scans. Each one takes longer. One or more of: `tables` (Tables (structure, headers and cells)), `text` (Text (misread words, missing text and numbers)), `formatting` (Formatting (bold, italic, sub and superscript)), `layout` (Layout (reading order and page structure)).
- Describe pictures and charts (`describe_figures`, true/false): Write a description of each picture and chart into the text.
- Cut the text into pieces (`chunk_types`, list): Also return the text cut into pieces, for search or AI steps: by meaning, by heading, by page or by size. One or more of: `semantic` (By meaning), `header` (By heading), `page` (By page), `recursive` (By size).
- Piece size (`chunk_size`, number): The most characters in one piece, 100 to 10,000. Used with 'Cut the text into pieces'. From 100 to 10000.
- Also return HTML (`include_html`, true/false): Return the document as HTML as well as Markdown.
- Folder in Pulse (`folder_name`, text): Save the extraction in this folder of the Pulse Platform's library. It is made if it doesn't exist.

**Outputs**

- `chunks` (object): The text cut into pieces, one list per way of cutting it that was asked for (semantic, header, page, recursive)
- `credits_used` (number): Credits this step used (empty when your plan doesn't use credits)
- `extraction_id` (text): The id of the saved extraction, for 'Extract fields', 'Split document', 'Extract tables' and 'Read charts'
- `extraction_url` (text): Where to see the extraction in the Pulse Platform
- `failed_pages` (list): Pages that could not be read, if any
- `html` (text): The document as HTML, when asked for
- `job_id` (text): The Pulse job that did the work
- `markdown` (text): The document's text, with headings and tables, as Markdown
- `page_count` (number): How many pages were read
- `warnings` (list): Notes from Pulse about the reading

**Example output**

```json
{
  "chunks": {
    "page": [
      "# Account statement",
      "Closing balance: $980.10"
    ]
  },
  "credits_used": 2.0,
  "extraction_id": "e1b2a3d4-0000-4000-8000-000000000002",
  "extraction_url": "https://platform.runpulse.com/dashboard/extractions/e1b2a3d4-0000-4000-8000-000000000002",
  "failed_pages": [],
  "html": "",
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "markdown": "# Account statement\n\nOpening balance: $1,204.50",
  "page_count": 2,
  "warnings": []
}
```

### Extract fields

Pull named fields (an invoice number, a total, a name) out of a document 'Extract document' has read, with where each was found and how sure Pulse is. Uses 1 credit a page.

**Inputs**

- Extraction ID (`extraction_id`, text, required): The extraction_id from 'Extract document' (or from 'Get job' once a started extraction has finished).
- Fields to extract (`fields`, object): Each field's name with what it is, for example {"invoice_number": "The invoice number", "total": "The amount due"}. Give this or a JSON Schema.
- JSON Schema (`json_schema`, object): Instead of 'Fields to extract', a JSON Schema of the result, for numbers, lists and nested fields.
- Instructions (`instructions`, text): Anything else Pulse should know when reading the fields.
- Take more care (`thorough`, true/false): Reason through hard documents more carefully. Uses 6 credits a page instead of 1.

**Outputs**

- `citations` (object): Where in the document each value was found, by field name
- `confidence` (object): How sure Pulse is of each value, from 0 to 1, by field name
- `credits_used` (number): Credits this step used (empty when your plan doesn't use credits)
- `job_id` (text): The Pulse job that did the work
- `schema_id` (text): The id Pulse saved these results under
- `values` (object): Each field's value, by field name

**Example output**

```json
{
  "citations": {
    "account_holder": [
      "txt-3"
    ],
    "closing_balance": [
      "tbl-1"
    ]
  },
  "confidence": {
    "account_holder": 0.98,
    "closing_balance": 0.95
  },
  "credits_used": 2.0,
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "schema_id": "5c1b2a3d-0000-4000-8000-000000000004",
  "values": {
    "account_holder": "Jo Bloggs",
    "closing_balance": "$980.10"
  }
}
```

### Extract fields from a section

Pull named fields out of only the pages 'Split document' found for one topic. Uses 1 credit a page of that section.

**Inputs**

- Split ID (`split_id`, text, required): The split_id from 'Split document'.
- Topic (`section`, text, required): The topic to read, named exactly as in 'Split document'.
- Fields to extract (`fields`, object): Each field's name with what it is, for example {"invoice_number": "The invoice number", "total": "The amount due"}. Give this or a JSON Schema.
- JSON Schema (`json_schema`, object): Instead of 'Fields to extract', a JSON Schema of the result, for numbers, lists and nested fields.
- Instructions (`instructions`, text): Anything else Pulse should know when reading the fields.
- Take more care (`thorough`, true/false): Reason through hard documents more carefully. Uses 6 credits a page instead of 1.

**Outputs**

- `citations` (object): Where in the document each value was found, by field name
- `confidence` (object): How sure Pulse is of each value, from 0 to 1, by field name
- `credits_used` (number): Credits this step used (empty when your plan doesn't use credits)
- `job_id` (text): The Pulse job that did the work
- `schema_id` (text): The id Pulse saved these results under
- `section` (text): The section the fields were read from
- `values` (object): Each field's value, by field name

**Example output**

```json
{
  "citations": {
    "account_holder": [
      "txt-3"
    ],
    "closing_balance": [
      "tbl-1"
    ]
  },
  "confidence": {
    "account_holder": 0.98,
    "closing_balance": 0.95
  },
  "credits_used": 1.0,
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "schema_id": "5c1b2a3d-0000-4000-8000-000000000005",
  "section": "Summary",
  "values": {
    "account_holder": "Jo Bloggs",
    "closing_balance": "$980.10"
  }
}
```

### Extract tables

Rebuild the tables of a read document, keeping merged cells and headers that span columns. Must be turned on for your organization by Pulse. Uses 0.25 credits a table.

**Inputs**

- Extraction ID (`extraction_id`, text, required): The extraction_id from 'Extract document' (or from 'Get job' once a started extraction has finished).
- Join tables split across pages (`merge_across_pages`, true/false): Join a table that runs onto the next page into one. Uses 0.5 credits a table.
- Turn charts into tables (`charts_to_tables`, true/false): Also make a table from each chart. 1 more credit a chart.
- Format (`table_format`, text): An HTML table, or a list of columns and rows. One of: `html` (HTML table (default)), `json` (Columns and rows). Default: `html`.

**Outputs**

- `credits_used` (number): Credits this step used (empty when your plan doesn't use credits)
- `job_id` (text): The Pulse job that did the work
- `table_count` (number): How many tables were found
- `tables` (list): The tables found
- `tables_id` (text): The id Pulse saved these tables under

**Example output**

```json
{
  "credits_used": 0.25,
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "table_count": 1,
  "tables": [
    {
      "citations": [
        "tbl-1"
      ],
      "from_chart": false,
      "table_content": "<table><tr><th>Date</th><th>Amount</th></tr></table>"
    }
  ],
  "tables_id": "7b1b2a3d-0000-4000-8000-000000000006"
}
```

### Fill form

Fill in a PDF form from plain-language instructions, on real form fields or on a flat scan, and save the filled PDF. Must be turned on for your organization by Pulse. Uses 3 credits a page.

**Inputs**

- PDF form (`file`, text, required): The PDF form: upload it, pick a file from an earlier step, or give a link Pulse can download.
- What to fill in (`instructions`, text, required): The details to put on the form, in plain words.
- Pages (`pages`, text): Only these pages, such as 1-3,5. Leave empty for the whole document.

**Outputs**

- `content_type` (text): The saved file's type (application/pdf)
- `credits_used` (number): Credits this step used (empty when your plan doesn't use credits)
- `document_ref` (text): The saved PDF, for a later step to attach
- `fields_filled` (number): How many fields were changed
- `filename` (text): The saved PDF's name
- `form_id` (text): The filled form's id in Pulse
- `job_id` (text): The Pulse job that did the work
- `page_count` (number): How many pages the PDF has
- `size_bytes` (number): The saved PDF's size in bytes
- `url` (text): Where to download the saved PDF in Lodol

**Example output**

```json
{
  "content_type": "application/pdf",
  "credits_used": 3.0,
  "document_ref": "doc_0123456789abcdef0123456789abcdef",
  "fields_filled": 7,
  "filename": "pulse-form-f1b2a3d4.pdf",
  "form_id": "f1b2a3d4-0000-4000-8000-000000000009",
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "page_count": 1,
  "size_bytes": 140815,
  "url": "/actions/core/serve-document/doc_0123456789abcdef0123456789abcdef"
}
```

### Get job

Read a Pulse job's status, and what it produced once it has finished. Free.

**Inputs**

- Job ID (`job_id`, text, required): The job_id from 'Start document extraction' or any other step.

**Outputs**

- `created_at` (text): When the job was started (ISO 8601)
- `error` (text): Why the job failed, when it did
- `extraction_id` (text): For a finished document reading, the saved extraction's id
- `is_finished` (true/false): True once the job has ended, however it ended
- `job_id` (text): The job's id
- `markdown` (text): For a finished document reading, the document's text
- `result` (object): Everything the finished job produced, as Pulse returns it
- `status` (text): pending, processing, completed, failed, canceled or expired
- `updated_at` (text): When the job last changed (ISO 8601)

**Example output**

```json
{
  "created_at": "2026-11-02T10:30:00Z",
  "error": "",
  "extraction_id": "e1b2a3d4-0000-4000-8000-000000000002",
  "is_finished": true,
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "markdown": "# Account statement",
  "result": {
    "extraction_id": "e1b2a3d4-0000-4000-8000-000000000002",
    "page_count": 2
  },
  "status": "completed",
  "updated_at": "2026-11-02T10:30:45Z"
}
```

### Get usage

Read your plan's credits used and left this billing period, and how many documents, pages and credits were used between two dates (the last 30 days by default). Free.

**Inputs**

- From (`from_date`, date): The first day to count. Leave empty for 30 days ago.
- To (`to_date`, date): The last day to count, included. Leave empty for today.
- Break down by (`group_by`, text): Count by day, week or month. One of: `day` (Day (default)), `week`, `month`. Default: `day`.

**Outputs**

- `billing` (object): Where the current billing period stands
- `by_period` (list): What was read in each day, week or month (empty ones left out)
- `from` (text): Start of the period counted (ISO 8601)
- `group_by` (text): How the period is broken down: day, week or month
- `to` (text): End of the period counted (ISO 8601)
- `totals` (object): What was read in the period counted

**Example output**

```json
{
  "billing": {
    "credit_allowance": 50000.0,
    "credits_remaining": 37548.5,
    "credits_used": 12451.5,
    "limit_exceeded": false,
    "period_end": "2026-12-01T00:00:00+00:00",
    "plan_name": "Growth"
  },
  "by_period": [
    {
      "credits": 803.0,
      "documents": 130,
      "pages": 708,
      "period": "2026-11-01T00:00:00+00:00"
    }
  ],
  "from": "2026-10-03T00:00:00+00:00",
  "group_by": "day",
  "to": "2026-11-02T23:59:59.999999+00:00",
  "totals": {
    "credits": 4184.0,
    "documents": 611,
    "pages": 3720
  }
}
```

### Read charts

Turn the line, bar, pie and scatter charts of a read document back into their numbers. Must be turned on for your organization by Pulse. Uses 1 credit a chart.

**Inputs**

- Extraction ID (`extraction_id`, text, required): The extraction_id from 'Extract document' (or from 'Get job' once a started extraction has finished).
- Pages (`pages`, text): Only charts on these pages, such as 1-3,5.
- Points per line (`data_points`, number): How many points to read along each line, 2 to 500. Pulse's default is 20. From 2 to 500.

**Outputs**

- `charts` (list): The charts, each with its data
- `charts_id` (text): The id Pulse saved these charts under
- `count` (number): How many charts were read
- `credits_used` (number): Credits this step used (empty when your plan doesn't use credits)
- `job_id` (text): The Pulse job that did the work

**Example output**

```json
{
  "charts": [
    {
      "confidence": 0.9,
      "id": "p2-chart-1",
      "page_number": 2,
      "series": [
        {
          "data": [
            [
              "Jan",
              410
            ]
          ],
          "name": "Spending"
        }
      ],
      "title": "Monthly spending",
      "type": "bar"
    }
  ],
  "charts_id": "8c1b2a3d-0000-4000-8000-000000000007",
  "count": 1,
  "credits_used": 1.0,
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001"
}
```

### Read form fields

Find the fields of a PDF form, filled in or blank, scanned or not, with what each one holds and which boxes are ticked. Must be turned on for your organization by Pulse. Uses 1 credit a page.

**Inputs**

- PDF form (`file`, text, required): The PDF form: upload it, pick a file from an earlier step, or give a link Pulse can download.
- Pages (`pages`, text): Only these pages, such as 1-3,5. Leave empty for the whole document.

**Outputs**

- `credits_used` (number): Credits this step used (empty when your plan doesn't use credits)
- `field_count` (number): How many fields were found
- `form_fields` (list): The fields found
- `form_id` (text): The form's id in Pulse
- `job_id` (text): The Pulse job that did the work
- `page_count` (number): How many pages the PDF has

**Example output**

```json
{
  "credits_used": 1.0,
  "field_count": 1,
  "form_fields": [
    {
      "checkbox_details": [
        {
          "selected": true,
          "text": "Individual"
        }
      ],
      "page_number": 1,
      "text": "Individual C corporation",
      "type": "checkbox"
    }
  ],
  "form_id": "f1b2a3d4-0000-4000-8000-000000000008",
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "page_count": 1
}
```

### Split document

Find which pages of a read document hold each topic you name, such as the financial statements of an annual report. Uses 2 credits a page.

**Inputs**

- Extraction ID (`extraction_id`, text, required): The extraction_id from 'Extract document' (or from 'Get job' once a started extraction has finished).
- Topics (`topics`, object, required): Each topic's name with what belongs in it, for example {"Financials": "Balance sheet and income statement"}.

**Outputs**

- `credits_used` (number): Credits this step used (empty when your plan doesn't use credits)
- `job_id` (text): The Pulse job that did the work
- `sections` (list): Each topic with its page numbers
- `split_id` (text): The split's id, for 'Extract fields from a section'
- `splits` (object): Each topic's page numbers, by topic name

**Example output**

```json
{
  "credits_used": 4.0,
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "sections": [
    {
      "name": "Summary",
      "pages": [
        1
      ]
    },
    {
      "name": "Transactions",
      "pages": [
        1,
        2
      ]
    }
  ],
  "split_id": "51b2a3d4-0000-4000-8000-000000000003",
  "splits": {
    "Summary": [
      1
    ],
    "Transactions": [
      1,
      2
    ]
  }
}
```

### Start document extraction

Start reading a document and go on at once, without waiting. Follow it with 'Get job'. For very long documents. Uses 1 credit a page.

**Inputs**

- File (`file`, text, required): The document: upload it, pick a file from an earlier step, or give a link Pulse can download. PDFs, images (JPG, PNG, TIFF and more), Word, PowerPoint, Excel, CSV and HTML files work. Uploads can be up to 100 MB.
- Pages (`pages`, text): Only these pages, such as 1-3,5. Leave empty for the whole document.
- About the document (`instructions`, text): Anything that helps Pulse read it, such as what kind of document it is or how to treat its numbers.
- Double-check (`refine`, list): Extra passes that correct what was read, for hard scans. Each one takes longer. One or more of: `tables` (Tables (structure, headers and cells)), `text` (Text (misread words, missing text and numbers)), `formatting` (Formatting (bold, italic, sub and superscript)), `layout` (Layout (reading order and page structure)).
- Describe pictures and charts (`describe_figures`, true/false): Write a description of each picture and chart into the text.
- Cut the text into pieces (`chunk_types`, list): Also return the text cut into pieces, for search or AI steps: by meaning, by heading, by page or by size. One or more of: `semantic` (By meaning), `header` (By heading), `page` (By page), `recursive` (By size).
- Piece size (`chunk_size`, number): The most characters in one piece, 100 to 10,000. Used with 'Cut the text into pieces'. From 100 to 10000.
- Also return HTML (`include_html`, true/false): Return the document as HTML as well as Markdown.
- Folder in Pulse (`folder_name`, text): Save the extraction in this folder of the Pulse Platform's library. It is made if it doesn't exist.

**Outputs**

- `job_id` (text): The job's id, for 'Get job' and 'Cancel job'
- `message` (text): Pulse's note about the job
- `status` (text): The job's status when started, usually pending

**Example output**

```json
{
  "job_id": "c1b2a3d4-0000-4000-8000-000000000001",
  "message": "Document processing started",
  "status": "pending"
}
```
