What You Will Need (Prerequisites)

Essential components and configurations needed before creating a RAG application in Dataworkz.

To get started with Dataworkz, you need a few essential components and configurations. The sections below cover each requirement.

Configuring LLMs in Dataworkz

Configure your Large Language Model (LLM) — the AI model responsible for reasoning and generating responses — before creating a RAG application in Dataworkz.

When you open the LLM Configuration section, you see a catalog of models you have already connected, along with details such as type, provider, and default status. From here, manage existing models or add a new one.

Adding a New LLM

Go to Configuration → LLM Configuration.

Click Add New Configuration to open the configuration panel. This is where you define the connection between Dataworkz and your model provider.

Choose the model type:

• Generative — answers questions and writes content. Most RAG setups use this type.

• Extractive — pulls structured values directly from documents.

After selecting the type, select the Deployment Type. Dataworkz supports multiple providers: OpenAI, Azure OpenAI, Amazon Bedrock, Gemini, or hosted models.

Selecting the Model

After choosing a provider, select from the dropdown of available models. For example, if you select OpenAI, options may include gpt-4, gpt-4o, or gpt-3.5-turbo. For Groq-hosted deployments, you might choose llama-3.3-70b-versatile.

Enter a friendly name (such as support_gpt4o_prod) to help your team identify the model later.

Paste in your API key. Dataworkz does not include built-in keys — you must provide your own from your LLM provider.

Fine-Tuning Behavior

Control how the model behaves using these settings:

• Max Tokens & Response Length — sets the context window size and output length. Use higher values when working with large enterprise documents to avoid truncating context.

• Temperature — controls creativity. Lower values (such as 0.1) produce factual, repeatable responses; higher values produce more exploratory ones.

• Penalties — adjust how much the model repeats itself or introduces new concepts.

If your use case requires image understanding (such as reading PDFs with diagrams), enable Visual Language Model (VLM). This tells Dataworkz that the model can process multimodal input.

Testing and Saving

Click Test Connection. This validates your API key and confirms the model is reachable. If the test succeeds, save the configuration.

Your new model appears in the list, where you can:

• Mark it as the default LLM.

• Edit parameters later if business needs change.

• Remove it if the key expires or the provider is no longer needed.

Example Setup: Customer Support RAG

A suitable LLM configuration for a RAG app handling customer policy questions:

• Model: gpt-4o (OpenAI, production)

• Temperature: 0.1 (ensures accurate, non-creative answers)

• Tokens: 8000 (supports large context from policy PDFs)

• Response length: 600 (keeps answers concise)

• VLM: Enabled (if policy guides include diagrams)

With this setup, when a customer asks "Can I return a damaged product bought last year?", the app retrieves the relevant rule from documents and answers reliably, without inventing details.

Storage Setup

Dataworkz provides flexible options for storing and managing data. Integrations for documents (RAG), structured enterprise databases, and graph data are available in the platform.

Default Storage

When you sign up, Dataworkz provides a default workspace with two built-in storage options:

• S3-compatible object storage — upload files such as PDFs, Word documents, or PowerPoint decks for RAG ingestion pipelines. Direct uploads from your local machine are limited to 1 MB per file. For larger uploads, connect your own external storage.

• MongoDB (Vector storage) — embeddings, text chunks, and metadata are stored here by default, acting as your vector database for semantic similarity search.

These defaults support quick experimentation, but most production projects will connect external enterprise-grade storage.

Supported Storage Options

From the Databases panel in Dataworkz, configure a vector database for storing and searching embeddings:

• Vector Databases: Pinecone, OpenSearch, Weaviate (coming soon)

Free-Tier Limitations

If you are on the free tier, keep the following limits in mind:

• File uploads are capped at 1 MB when done directly from your machine.

• The web crawler supports up to three HTML pages per month.

• Vector storage is provisioned via MongoDB only. External database connections are available on paid tiers.

Data Privacy and Best Practices

The default workspace is convenient for getting started. Before uploading sensitive or proprietary data, set up your own dedicated vector and object storage (such as S3, GCS, or your enterprise database of choice).

Once external storage is configured, you can:

• Migrate existing datasets out of the default workspace.

• Adjust workspace and collection settings for stricter access control.

• Ensure compliance with internal data governance policies.