Agentic parse to markdown

Nodes used: Parse (mode: "agentic") The Agentic tier is the same Parse node with mode set to agentic. It parses in several steps and spends the most work on dense tables, so it is the tier to reach for when the Standard output on a complex layout is not good enough.

The graph

{
  "name": "Agentic parser",
  "nodes": [{ "id": "parse_1", "type": "parse", "mode": "agentic", "effort": "mid" }],
  "edges": []
}

import os
from anyformat.sdk import Client

client = Client(api_key=os.environ["ANYFORMAT_API_KEY"])

workflow = client.workflow("Agentic parser").parse(mode="agentic", effort="mid").create()

import { Anyformat } from "@anyformat/sdk";

const af = new Anyformat({ apiKey: process.env.ANYFORMAT_API_KEY! });

const workflow = await af.workflow("Agentic parser").parse({ mode: "agentic", effort: "mid" }).create();

effort is an Agentic-only knob:

`effort`	Trade-off
`low`	Several times cheaper. More errors on dense or low-contrast tables.
`mid`	The default. Balanced.
`accurate`	Highest fidelity. Slowest and most expensive.

prompt_hint also applies, for example "the second column is a date". figure_enhancement is ignored on this tier.

Run and read

Agentic runs are slower than Standard: 37 s for a one-page invoice against 28 s, and minutes for a long document. Give wait() a bigger budget.

curl -X POST "https://api.anyformat.ai/v3/workflows/$WORKFLOW_ID/upload/run/" \
  -H "Authorization: Bearer $ANYFORMAT_API_KEY" \
  -F 'files=@report.pdf'
# -> 202 {"run_id": "...", "status": "queued"}

# Read the run. Repeat until "status" is "processed".
curl "https://api.anyformat.ai/v3/runs/$RUN_ID/" \
  -H "Authorization: Bearer $ANYFORMAT_API_KEY"

result = workflow.run("report.pdf").wait(timeout=600)

print(result.parse.parse_confidence)
for block in result.parse.blocks:
    if block.type == "table":
        print(block.id, block.parse_confidence, len(block.rows or []), "rows")

import { readFile } from "node:fs/promises";

const file = new File([await readFile("report.pdf")], "report.pdf", { type: "application/pdf" });
const run = await workflow.run(file);
const result = await run.wait({ timeoutMs: 600_000 });

console.log(result.parse?.parseConfidence);
for (const block of result.parse?.blocks ?? []) {
  if (block.type === "table") console.log(block.id, block.parse_confidence, block.rows?.length ?? 0, "rows");
}

What comes back

The Agentic tier on a one-page invoice, trimmed. The shape is the same as every other tier; only the work behind it differs.

{
  "id": "06a8d903-9bb2-70e3-8000-c15378b47b7c",
  "status": "processed",
  "results": {
    "document_packet_id": "06a8d903-9a05-7184-8000-d34b711c8ef0",
    "verification_url": "https://app.anyformat.ai/workflows/.../files/...",
    "parse": {
      "markdown": "<a id=\"p1_b0\"></a>\n\nINVOICE\n\n<a id=\"p1_b1\"></a>\n\nInvoice #: INV-2026-9001\nIssue date: 2026-02-20\n...\n\n<a id=\"p1_b5\"></a>\n\n<table><thead><tr><th data-cell-id=\"r0c0\">Description</th><th data-cell-id=\"r0c1\">Qty</th> ...",
      "text": "INVOICE\nInvoice #: INV-2026-9001 ...",
      "parse_confidence": 75.0,
      "layout_confidence": 0.6,
      "blocks": [
        { "id": "p1_b0", "type": "header", "page": 1, "parse_confidence": 100.0, "content": "INVOICE", "rows": null },
        { "id": "p1_b5", "type": "table", "page": 1, "parse_confidence": 85.9, "content": "<table>...</table>",
          "rows": [
            [{ "cell_id": "r0c0", "text": "Description" }, { "cell_id": "r0c1", "text": "Qty" }, { "cell_id": "r0c2", "text": "Unit price" }, { "cell_id": "r0c3", "text": "Amount" }],
            [{ "cell_id": "r1c0", "text": "Matrix Extensible Action-Items" }, { "cell_id": "r1c1", "text": "8" }, { "cell_id": "r1c2", "text": "$246.87" }, { "cell_id": "r1c3", "text": "$1,974.96" }]
          ] },
        { "id": "p1_b6", "type": "table", "page": 1, "parse_confidence": 31.0, "content": "<table>...</table>", "rows": [ "..." ] }
      ]
    },
    "classifications": [],
    "splits": [],
    "extractions": [],
    "edits": []
  }
}

Reading the confidence

blocks[].parse_confidence is the number to act on. It is 0 to 100 per block. In the sample above the line-item table scored 86 and the totals block, which the tier read as a second table, scored 31. That is the block to send to review.
parse.parse_confidence is the document roll-up, weighted by the characters in each block. A long low-scoring table pulls it down more than a short heading pulls it up.
A block that no model scored comes back with parse_confidence: null. Treat null as “not scored”, not as zero.

Tips

Try Standard first. Move a document type to Agentic only when its tables come back wrong. The Parse node page lists the credits per page for each tier.
Start at effort: "mid". Go to accurate for scanned, low-contrast or very wide tables; drop to low for clean digital documents where you want the multi-step pass without the cost.
rows on a table block gives you the cells as a grid. Read that instead of parsing the <table> HTML.
Add an Extract node after the Parse node to turn the same tables into typed fields. See Bank statement processing.

Next steps

Parse node

Every knob on the node

Parse-only workflow

The same graph on Flash, Fast and Standard

Get run

The full run envelope

Coding assistant

The anyformat skill for Claude Code and other agents

Examples

Read only

Read, then pull out

Sort first, then pull out

Extract, then check

Operate at scale

Agentic parse to markdown

The graph

Run and read

What comes back

Reading the confidence

Tips

Next steps

Parse node

Parse-only workflow

Get run

Coding assistant

​The graph

​Run and read

​What comes back

​Reading the confidence

​Tips

​Next steps

Parse node

Parse-only workflow

Get run

Coding assistant

The graph

Run and read

What comes back

Reading the confidence

Tips

Next steps