Datasets – Running AI Tasks on Files

Datasets are the workspace where PromptLoop stores every spreadsheet you upload and every version created after an AI task or generation run. Upload a CSV or Excel file, pick the columns you want the model to use, and run tasks on hundreds—or thousands—of rows in one go.

PromptLoop handles all the heavy lifting: queuing, parallel execution, and versioning. You just download the enriched file when it’s done.

Helpful links  

• Datasets Page  
• Generate a Dataset  
• Create a Task

Getting Started (2‑minute video)#

Watch the video

Working with Files & Versions#

Uploading a Spreadsheet#

1. Go to Datasets → Upload File.
2. Drag‑and‑drop a .csv or .xlsx file.
3. PromptLoop auto‑detects columns. Tick the one(s) you plan to use as task inputs (often a single column such as website or search_term).
4. Click Save Dataset.

Tip: You can upload a file just to explore and filter it first—there’s no need to run a task immediately.

Launching an AI Task#

1. On the dataset row click Launch Job.
2. Choose an existing task or click Create New Task if you haven’t built one yet (the editor opens in a new tab).
3. Map the dataset column(s) to the task input(s).
4. Hit Run. Jobs queue instantly; even large files (20 k–50 k rows) usually finish within 90 minutes.

Reviewing Results#

When the job completes PromptLoop writes a new version of the dataset.

• Use the in‑app filters and search bar to spot errors or blanks quickly.
• Download a fresh CSV/XLSX or chain another task right away.

Results stay in your account indefinitely unless you delete the dataset.

Dataset Tools#

Generating a Dataset (Autoloop)#

Dataset Generation lets you build a prospect list from scratch using PromptLoop’s crawlers—no upload required.

Generation Modes#

How to Generate#

1. Open Datasets and click Generate Dataset.
2. Pick Geographic or Global.
3. Fill in the form → Generate.
4. A placeholder row appears while the crawler works (usually a few minutes). The row flips to Completed when done.
5. Treat the new dataset like any other—run enrichment tasks or download it.

New Column Gen (Excel‑style quick transforms)#

Use New Column Gen to reformat or combine existing columns without leaving PromptLoop.

1. Click New Column on any dataset.
2. Select the columns to feed into the prompt.
3. Describe what you want (“Return the domain part after https://”).
4. Preview on random rows until output looks right.
5. Click Generate → PromptLoop adds a new version with the extra column.
   Speed guide: ~10 minutes per 20 k rows.

Best Practices & Limits#

• Column hygiene: keep input columns clean (one URL per cell, no trailing text).
• Batch size: for >50 k rows split into chunks; run two jobs back‑to‑back to avoid browser timeouts.
• Version control: rename important versions and save a filter or view that you need for a new task.

FAQ#

Can I delete intermediate versions?
Yes. Click the kebab (⋮) on the version row → Delete.

How do I rerun a task only on blank rows?
Filter the dataset where the target column is empty, export those rows as a new file, re‑upload, and run the task.

What file types can I upload?
.csv, .xls, .xlsx. TSV works if you rename to .csv first.