Assistant Batch Processing

Deutsch: Assistent Stapelverarbeitung

The "Batch Processing" assistant applies the same instructions to every document of a folder, one document at a time, and collects the answers: as one file per document, or as one table with a row per document. Use it for many documents of the same kind, such as applications, reports, or recordings.

How to use it

1. Open "Assistants" → "Batch Processing".
2. Under "Input":
   • "Folder containing your documents": choose the folder, or drop it onto the field.
   • "Include subfolders?": also process the files in subfolders.
   • "Files to include": which files to process. The default covers PDF, Word, PowerPoint, Excel, Markdown, and text files, images, and audio and video files. Separate patterns with semicolons, for example, *.pdf;*.docx.
3. Under "Instructions", choose the "Source of the instructions":
   • "Use a free prompt": describe under "What should the AI do with each document?" what the AI should do with each document.
   • "Use a document analysis policy": use a policy from the Document Analysis assistant, with its rules, tools, minimum confidence, and provider.
   • "Import from a file (.md)": use the content of a Markdown file as instructions.
4. Under "Output", choose the "Output mode"; see below.
5. Choose a provider, and click "Start batch processing".

With a free prompt or a file, "Tools for this batch run" lets the AI use tools for each document.

Output

• "One file per document": each answer is stored as a file named after its document, for example, report_result.md for report.pdf. Choose the "File format": Markdown, or Word, OpenDocument, LaTeX, or HTML, which need Pandoc; see Pandoc and File Formats.
• "One CSV results table, where each answer becomes one row": one table with the file name and the answer of each document. You can name the table and the column, and choose the "Column separator". With this mode, ask the AI for short answers.

By default, AI Studio writes the results into the subfolder ai-results of your input folder. Choose another one under "Output folder (optional)".

In addition, AI Studio always writes log.csv, which lists every document with the time it finished, the model, the status, and any error.

While it runs

A table shows each file as "Queued", "Done", "Failed", or "Canceled", and a progress bar counts the processed files. Between two files, AI Studio waits a random number of seconds, by default 6 to 10. You set this pause under "Processing pace".

If a file fails, the run continues with the next one; the error appears in the table and the log. "Cancel the batch run" stops the run; the remaining files stay queued, and you can continue the run later.

You can leave the assistant while it runs. If AI Studio is closed during a run, the results so far stay on disk.

Continuing a run

When you start a run on an output folder that already has a log, AI Studio asks "Continue the previous batch run?":

• "Continue the previous run" processes only the documents that failed or are missing.
• "Start a new run" processes all documents again.

Things to know

• Costs: each document is a request to your provider, so a large folder can cost money. Try your instructions on a small folder first.
• Recordings: audio and video files are transcribed by your transcription provider without asking again. Each transcript is stored next to its file as <file>.transcript.md and reused when you continue a run.
• Images are given to the model as images, so it must be able to read them. Otherwise, remove the image patterns from "Files to include".
• Expert settings: "Show Expert Settings" reveals hidden files, excluded files and folders such as node_modules/, and a size limit of 1,024 MB by default.

Options

In "Assistant: Batch Processing defaults", "Preselect batch processing options?" lets every new run start with your defaults: input and output folder, file patterns, instructions, output mode, file format, table settings, pause, and provider.

Where to go next

• Assistants
• Document Analysis