Knowledge sources
Files and images
Upload documents, spreadsheets and images to an Internal Files connection, and review how Harmona reads each one.
La documentación está en inglés y turco, como la interfaz de Harmona. Esta página se muestra en inglés.
An Internal Files connection holds files you upload from your computer. Agents can read every file in full, search across them and cite the file they used.
Upload files
- 1Go to Studio → Connections and click Create Connection.
- 2Choose Internal Files.
- 3Enter a Connection Name and, optionally, a Description. A clear description helps agents pick the right source.
- 4Drag files onto the upload area, or click to browse. You can also skip this and add files later.
- 5Click Create Connection. Processing starts right away.
To add more files later, open the connection, select the source and click Add Files. Files in an unsupported format are skipped, and Harmona tells you how many.
Consejo
Supported formats
| Kind | Formats | What Harmona does |
|---|---|---|
| Documents | PDF, DOC, DOCX, TXT, MD | Extracts the text and indexes it for search |
| Spreadsheets | CSV, XLS, XLSX, XLSM | Loads each CSV file and each Excel sheet into a SQL table |
| Parquet | PARQUET | Queries the file in place as a dataset; files with the same columns form one table |
| Images | PNG, JPG, JPEG, WEBP, HEIC, HEIF | Reads the image with a vision model into a searchable description |
Processing time depends on file size. Each file shows a status: Pending, In Progress, Completed or Failed.
Spreadsheets become tables
Every CSV file, Excel sheet and Parquet dataset in a connection becomes a table agents can query with SQL. So a question like "total sales by region across all files" is answered in one query, not by re-reading files.
After indexing, an analysis writes a card for each table: what it holds, what each column means and how tables join. Open Document Processing on the connection and go to Spreadsheet tables to read or edit the cards. Click Describe tables to write missing cards, or Reindex to reload every spreadsheet and describe the tables again.
Image recognition
Photographed menus, scanned invoices and screenshots become answerable too. A vision model transcribes the text in the image and describes what it shows. The description is indexed like any document.
This works for uploaded images and for images that arrive through a synced source, such as SFTP, SharePoint or Google Drive.
Each image shows a review state:
| State | Meaning |
|---|---|
| Not analyzed | No description yet. Auto recognition was off, or the image has not been analyzed. |
| Auto analyzed | The vision model wrote the description. Nobody has reviewed it yet. |
| Controlled | A person edited or approved the description. |
Open an image to see it next to its description. Click Analyze to run the vision model on demand, correct the text and click Save information, or click Approve to confirm the description as it is.
Consejo
Automatic or on demand
The Auto image recognition setting under Management → Organization Settings → General decides whether new images are analyzed as they arrive. It is on by default. Turn it off if you prefer to control cost or review images first. Images then stay Not analyzed until you click Analyze on them.
Actualizado 2026-09-24