Skip to main content
A VirtualBrain AI App is a multi-block pipeline. Each block is a separate call to the Mainframe that produces a text deliverable, such as a report, a scoring grid, a memo, or a comparison table. Blocks run in sequence or in parallel, and the output of one block can feed the next. Each block runs on an agentic orchestrator. During a run, a block can internally trigger specialized tools and iterate until it has what it needs:
  • Document Retriever, which searches across Brains and uploaded documents.
  • Web Search, for live external context.
  • Spreadsheet/Excel Analysis, which runs real code over Excel files, CSVs, and tables inside PDFs.
  • Document Comparison, which reports what changed between two versions of a document.
The output of a block remains text or tables; the tools are orchestrated invisibly during the run. Several Knowledge Sources, including Web Search, can now be combined within the same block.

​
The three things that decide quality

Get these three right and the App holds up across every run.
1

Architecture

The right blocks, each with exactly one objective. A block that tries to do two jobs produces a weaker result on both.
2

Data flow

How inputs, documents, and intermediate results reach each block.
3

Prompts

Short, structured, and focused. See App Blocks for how a prompt and its Knowledge Sources combine inside one block.
Data flow is the #1 source of broken Apps. How content reaches each block, injected directly into the prompt or retrieved from a Knowledge Source, decides whether the App is reliable. Passing Data Safely Between Blocks is the section that gets this right.
Think of the Mainframe as a senior partner delegating to specialists: you set the objective and hand over the right materials, and the tools do the retrieval, search, and analysis underneath. Your job as the App designer is to decide what each block works from and how the results travel between them. See Building Apps for the builder where you configure all of this.

​
Next

Key Concepts and Vocabulary

The shared vocabulary every prompt and design discussion relies on.