DeepRAG

The DeepRAG tool enables agents to analyze and synthesize information across large PDF and text files using retrieval-augmented generation (RAG). It is designed for scenarios where you need comprehensive, citation-backed answers derived from long-form content such as reports, manuals, policies, or research documents.

DeepRAG ingests one or more documents provided at runtime and performs semantic retrieval across their contents. Based on the task you define, it synthesizes relevant information into a coherent response and can optionally include citations that reference the source documents.

Typical use cases include:

• Summarizing large documents or collections of documents
• Answering complex questions grounded in source material
• Comparing or synthesizing information across multiple sections or files
• Producing research-style outputs with traceable citations

Data sources

DeepRAG supports just-in-time (JIT) file inputs.

• Files are provided directly through agent inputs.
• Only PDF and text files are supported.
• No index creation or external storage setup is required.

Configuring the DeepRAG tool

Add the DeepRAG tool

1. Open your agent in design mode.
2. From the built-in tools list, add DeepRAG to the agent.

Once added, the DeepRAG tool appears as a dedicated tool node that you can configure from the right-side panel.

Configure inputs

At design time, you define how DeepRAG should analyze the provided documents:

• Source file: Select the file or document input that DeepRAG should analyze. At runtime, this resolves to the document provided when the agent runs.s.

  Note:

  You must configure a file type argument in Data manager.

• DeepRAG task - Describe what the agent should research across the documents. This typically includes:

  • What information to look for
  • How to synthesize or structure the output
  • Whether and how to reference sources

• File extension - Specify the expected file type. This helps DeepRAG apply the appropriate document parsing strategy.

• Enable citations - When enabled, the tool includes citations in the output that reference the source documents used during synthesis.

  

Running the DeepRAG tool

When you run or debug the agent, DeepRAG executes as part of the agent run:

• The input document is loaded and parsed.
• Relevant sections are retrieved based on the task definition.
• The model synthesizes a response grounded in the retrieved content.
• An output is generated, optionally including citations.

During execution, DeepRAG appears in the Execution Trace as a dedicated tool call. You can expand this step to inspect detailed runtime information.

Inspecting execution details

In the Execution Trace, the DeepRAG tool appears as a dedicated tool call. You can expand it to review:

• Input - The source document, including file name, type, and metadata, and the query used to guide the retrieval and synthesis.
• Output - The generated response, including an array citations when enabled.

This visibility helps you understand how the answer was produced and which source content contributed to the final output.