Manycrawl

Help centre › Enrichment

Classify: yes or no, pick one, or score

Use Classify when every row needs the same kind of answer: yes or no, one option from a list, or a score. Each answer comes with a confidence value, so you can filter down to the rows the model was sure about. It runs on TypeSafe’s Jev model with your own TypeSafe key.

Before you start

Add your TypeSafe key in Settings → AI LLM models, under TypeSafe (Jev). Keys come from console.typesafe.ai, under Settings → Keys. You can pick Jev (latest), Jev (preview) or Jev 1.13.0 (pinned). See Which API keys do I need?

Add the column

  1. Click Enrichment and choose Classify.
  2. Write your question under Question for Jev. Type / to insert the column it should read, for example Body text. The question must refer to at least one column, otherwise Jev would judge every row on the question alone and give them all the same answer.
  3. Choose the Question type (Jev):
    • Yes / no: the question is a statement about the row, such as “This business is a dental clinic.” The cell shows Yes or No.
    • Pick one: one option out of a list. Under Options, write one option per line as value: what it means. The cell gets the value on the left of the colon. You need at least two options.
    • Score: rate against a scale. Under Options, write one level per line, lowest first, at least two. The cell gets a number with two decimals.
  4. Leave Add a confidence column ticked (it is on by default).
  5. Set Parallel rows (1–8). The default is 4. Lower it if your key starts hitting rate limits; 1 runs one row at a time.

Use the confidence column

The extra column is named after your column, with “confidence” at the end, and holds a number from 0 to 1. For a yes/no answer it is always at least 0.5, because it shows how sure the model is of whichever answer it gave. Filter it to keep only confident rows, for example confidence above 0.9, and review the rest by hand. See Control what a run costs.

If a row shows an error

MessageWhat to do
“TypeSafe API key not found. Please configure it in Settings → AI LLM models.”Add your key.
“Invalid TypeSafe API key.”Check the key in Settings.
“TypeSafe rate limit reached. It retries on its own.”Lower Parallel rows.
“TypeSafe is overloaded (HTTP 529). It retries on its own.”Wait. If it persists, run the failed rows later.
“TypeSafe rejected the question”Reword the question or check the options.
“Add at least two options…” or “…two score levels…”Add more lines under Options.
“Classify can’t run on model…”Classify needs a TypeSafe (Jev) model.

You pay TypeSafe directly for the tokens, and each row also counts as a Manycrawl action on the Free plan.