Bulk AI Prompt: ChatGPT-style LLM on Every Row
Pricing
from $2.00 / 1,000 row processeds
Run one AI prompt on every row of a CSV, Excel file, Google Sheet or dataset: classify, extract fields, summarise, write lines. Answers as new columns. All-inclusive price, no API key.
The instruction the AI model gets for each row. Put column names in double curly braces, e.g. {{review}} or {{product_name}}; they are replaced by the row's values. Names are not case-sensitive, and {{address.city}} reaches into nested values. Without any {{placeholder}}, the whole row is added to the prompt as JSON.
Ask the model for exactly these fields, e.g. sentiment, reason, category. Each field becomes its own column ai_ (missing values are empty). Leave empty to get one plain-text answer in the column ai_answer.
Your data as a JSON array of objects, one object per row, e.g. [{"review": "Arrived fast, works well"}]. Handy for API calls and quick tests. Not used when you also pick a file link or a dataset.
A public link to a .csv, .tsv or .xlsx file (first sheet, first row = column names), or a Google Sheets link shared as "Anyone with the link" (the tab in the link is used). Google Drive, Dropbox and OneDrive share links work too. Max. 50 MB.
Use the items of a dataset as rows, e.g. the output of another actor. The actor only gets read access to the dataset you pick.
General instructions that apply to every row, e.g. "You are a careful e-commerce analyst. Answer in English." Leave empty for a short default that asks for the answer only.
Standard is a fast mid-size model, fine for classifying, extracting and short texts. Large is a 70B model for harder instructions. Both run on Cloudflare Workers AI; the price per row includes the model.
Only the first rows up to this number are read and processed.
A filled-in prompt longer than this is shortened: the longest values are cut first and end in " [...]", so your instruction stays intact. The run summary counts how often this happened.
Upper limit for the model's answer per row (a token is roughly 3/4 of a word). Raise it for longer texts.
0 = the most predictable answers (best for classifying and extracting), higher = more varied wording (for creative lines).
