Tabular Data

Agentset processes tabular data so that rows stay intact and headers remain associated with their values. This makes spreadsheets, CSVs, and tables embedded in documents searchable without losing structure.

Formats

Agentset detects tabular data in two ways:

Method	Description
Spreadsheet files	Upload CSV, Excel (`.xls`, `.xlsx`), or OpenDocument (`.ods`) files directly.
Tables in documents	Tables embedded in PDFs, Word docs, and presentations are automatically detected and extracted.

Both methods preserve table structure automatically—no configuration required.

Why table-aware processing matters

Standard text chunking breaks tables at arbitrary points, separating headers from their data and splitting rows mid-content. This destroys the relationships that make tabular data meaningful. Consider a product inventory table:

Product	SKU	Price	Stock
Wireless Mouse	WM-001	$29.99	150
USB Keyboard	KB-002	$49.99	89
Monitor Stand	MS-003	$79.99	34
Webcam HD	WC-004	$89.99	67
USB Hub	UH-005	$24.99	203
Laptop Stand	LS-006	$59.99	56

Without table-aware processing, a chunk boundary might fall between rows 3 and 4. Headers are lost from the second chunk.

Chunk 1:
"Product SKU Price Stock Wireless Mouse WM-001 $29.99 150 USB Keyboard
KB-002 $49.99 89 Monitor Stand MS-003 $79.99"

Chunk 2:
"34 Webcam HD WC-004 $89.99 67 USB Hub UH-005 $24.99 203 Laptop Stand
LS-006 $59.99 56"

With Agentset, chunks are generated as a markdown table, always containing the full header and are not split mid-row.

Chunk 1:
| Product        | SKU    | Price  | Stock |
|----------------|--------|--------|-------|
| Wireless Mouse | WM-001 | $29.99 | 150   |
| USB Keyboard   | KB-002 | $49.99 | 89    |
| Monitor Stand  | MS-003 | $79.99 | 34    |

Chunk 2:
| Product       | SKU    | Price  | Stock |
|---------------|--------|--------|-------|
| Webcam HD     | WC-004 | $89.99 | 67    |
| USB Hub       | UH-005 | $24.99 | 203   |
| Laptop Stand  | LS-006 | $59.99 | 56    |

A search for “laptop accessories under $50” can now match relevant rows because each row contains the full context needed for retrieval.

Upload a spreadsheet

import { Agentset } from "agentset";

const agentset = new Agentset({
  apiKey: process.env.AGENTSET_API_KEY,
});

const ns = agentset.namespace("YOUR_NAMESPACE_ID");

const job = await ns.ingestion.create({
  payload: {
    type: "FILE",
    fileUrl: "https://example.com/inventory.csv",
  },
});

console.log(`Upload started: ${job.id}`);

import os
from agentset import Agentset

client = Agentset(
    namespace_id="YOUR_NAMESPACE_ID",
    token=os.environ["AGENTSET_API_KEY"],
)

job = client.ingest_jobs.create(
    payload={
        "type": "FILE",
        "fileUrl": "https://example.com/inventory.csv",
    }
)

print(f"Upload started: {job.data.id}")

Tables embedded in PDFs, Word documents, and presentations are processed the same way—upload the file and Agentset handles the rest. See File Uploads for uploading local files and other options.

Tables are automatically detected and processed. No additional configuration is required.

Next steps

API Reference — Tabular data ingestion parameters
File Uploads — Learn about supported file types and upload methods
Document Metadata — Attach metadata for filtering
Search — Query your tabular content

Get Started

Data Ingestion

Search and Retrieval

Production

Webhooks

Cookbooks

Formats

Why table-aware processing matters

Upload a spreadsheet

Next steps

​Formats

​Why table-aware processing matters

​Upload a spreadsheet

​Next steps

Formats

Why table-aware processing matters

Upload a spreadsheet

Next steps