Quickstart

Get started with Batch4LLM. This guide shows you how to extract the title from a set of PDF documents.

What you need:

• A running Batch4LLM instance (local or hosted)
• PDF documents to process
• A Gemini API key (this guide uses Gemini, but all providers work)

First batch?

If this is your first time, follow the steps below. Start with no more than 10 documents so you can see results quickly.

1. Create an endpoint

   Caution

   Your API key cannot be seen by other users in the frontend. However, your administrator and anyone with backend access can see it. Make sure you trust your platform. Set a spending limit on your key before using it.

   An endpoint connects Batch4LLM to an LLM provider. To create one, in the Sidebar on the left, click Endpoints. Then click Add Endpoint in the top right corner.

   In the pop-up, select Gemini as the provider. Enter "gemini" as the name and paste your Gemini API key. Click Add Endpoint. The endpoint is validated in the backend before it’s added. If there’s a problem with the API key or the endpoint, the error is shown immediately (see Endpoint connection errors).

2. Create a prompt

   A prompt is the instruction sent to the model for each document. In the Sidebar on the left, click Prompts. Then click Add Prompt in the top right corner.

   You’ll be asked to choose between Simple and Complex. Choose Simple — it sends your text as the prompt exactly as you enter it. Complex can be used later to build prompts made up of multiple sub-tasks, giving the LLM instructions in several separate steps.

   Enter "Retrieve pdf title" as the name. Paste the following into the text box:

   Your given a pdf document. Please extract the title from it.
   Only answer with the document title without
   any formatting or additional information.

   Leave all other settings as they are and click Add. Your new prompt will now appear in the overview.

3. Upload your documents

   Go to Files and click Upload Files. Either click the file box to browse your files, or drag and drop them in. Your files will appear in a list below the box.

   Type My documents in the file tag field and press Enter. Then click Upload files. Uploading may take a few seconds or a few minutes depending on how many files you have.

4. Create a batch

   A batch processes your documents using your chosen endpoint and prompt. Go to Batches in the Sidebar and click Create Batch.

   In the pop-up, click on Select a model.... For this guide, select gemini-2.5-flash. You can also use the search bar to look for a specific model.

   Then click on Select files..., click the My documents tag you assigned earlier, and click Done in the bottom right corner.

   Next, select the prompt you just created. For the file handling, leave the setting on Upload. This uploads your documents as they are, and the model provider processes them directly on their end.

   Caution

   Upload is not available for all endpoints and file types. For example, DeepSeek only supports Upload for images. If Upload does not work for your endpoint or file type, select a File Reader instead.

   Leave all other settings as they are and click Start Batch. Wait for the batch to finish.

   What happens in the backend?

   With Upload, the backend sends each document directly to the model provider, together with your chosen prompt. With a File Reader, the backend extracts the text from each document first and sends that instead. Either way, one API request is sent per document, and each document is processed in its own context window. This means you cannot use the prompt to compare or combine multiple documents.

   When the batch is done, you will see a green indicator. Click on the batch and select View details to get an overview of all the files processed, the LLM answers, and the token and cost usage.

5. Export your results

   Return to the Batch Overview by clicking the arrow in the top left corner. Then, in the same column as View Details, select Export on your finished batch.

   Leave the settings in the pop-up as they are and click Export. Allow downloads in your browser if prompted.

   You will receive a CSV file with one row per document. The LLM answer — in this case, the PDF title — is included as its own column.

You’re done! You just processed a set of documents with a prompt and exported the results.