FAQ¶
Installation and authentication¶
ImportError asking you to install a package¶
The chosen provider needs the corresponding extra. The message says which one to install, for example:
The extras are listed in Installation. LangChain providers without their own extra, such as Cohere and Mistral, need the langchain-<provider> package.
Authentication error (401 or 403)¶
The API key was not found or was rejected. Check the provider's environment variable (OPENAI_API_KEY, GOOGLE_API_KEY, ANTHROPIC_API_KEY, GROQ_API_KEY) in the same process that runs Python; in a notebook, set it before calling dataframeit(). An authentication error gets no new attempt: every row ends with status 'error'.
The codex and claude_code providers use the local authentication of those tools and do not accept api_key; see Providers.
Result¶
KeyError: '_dataframeit_status'¶
When no row fails and none records any detail, the status column and _error_details are removed from the output. Check that the column exists before filtering:
if '_dataframeit_status' in result.columns:
errors = result[result['_dataframeit_status'] == 'error']
ValueError about the text column¶
Without text_column, dataframeit looks for the columns texto, text, decisao, content and content_text, in that order, or uses the DataFrame's only column. With several columns and none of these names, set the column:
Some rows ended up with "Texto ausente"¶
Rows with empty or null text are not sent to the model. They get status 'error' and this detail, so that they do not pass as processed.
I ran it again and the rows with errors were not redone¶
With resume=True (default), only rows without status are processed. To redo the ones that failed, clear their status; see Reprocessing Errors.
The model columns already exist and nothing was processed¶
With resume=False, if the model columns are already in the DataFrame and reprocess_columns was not passed, dataframeit emits a warning and returns the data unprocessed, so as not to overwrite results. Use resume=True to continue, or reprocess_columns=[...] to redo specific columns.
Limits and costs¶
Many 429 errors (rate limit)¶
The provider rejected requests for exceeding its rate. dataframeit already halves the workers on each 429 and retries with backoff, but the stable fix is to lower parallel_requests or raise rate_limit_delay. The calculation is in Performance. With web search, the search provider's limit is usually the tightest; see Web Search.
How much will it cost?¶
Run a sample first (df.sample(30)) with track_tokens=True. The summary at the end of the run shows the tokens and, with search, the credits spent; multiply by the ratio between the dataset size and the sample size. Prices per model are in Providers.
A long run was interrupted¶
Use batch_size and checkpoint_path to save progress during the run, and resume by running the same call again, which continues from the file; see Checkpoints for Long Runs.