Skip to content

Lens Task Agent

The Lens Task Agent is an AI-powered vision and document analysis agent. It is designed to extract technical evidence, text, tables, and timing diagrams from images and documents.

It can be used directly via the CLI to analyze files on the fly, or it can be invoked by manager agents (such as the ASIC or Design agents) when they need to "see" a screenshot, datasheet figure, or schematic.

Key Features

  • Multi-Format Vision: Processes standard images (.png, .jpg, .jpeg, .gif, .webp, .bmp) and automatically renders .pdf files into analyze-able images.
  • Technical Extraction: Specifically tuned to interpret datasheets, schematic diagrams, timing diagrams, and tables, returning structured summaries, key elements, and deep technical details.
  • Targeted Analysis: Supports optional prompting to focus the analysis on specific signals, tables, or sections of a document.
  • Structured Output: Can emit JSON for seamless integration into verification or specification-generation pipelines.

Command Options

The Lens task agent supports the following command-line options:

Option Description
--input-image Required. Path to the image or PDF file to be analyzed.
--image-prompt Optional context or question to guide the analysis (e.g., "Extract the SPI timing parameters").
--json-output Emits structured JSON output containing the description, key elements, and technical details.

Usage

Use the Lens task agent through the NEX CLI by specifying --agent=lens.

Basic Command

Analyze a screenshot of a timing diagram:

nex --agent=lens --input-image=docs/timing_diagram.png

Guided Analysis

Focus the extraction on a specific detail within a datasheet PDF:

nex --agent=lens --input-image=datasheets/sensor_spec.pdf --image-prompt="Extract the register map table from this document"

JSON Output (For Tooling)

If you are using Lens in a script to populate a specification:

nex --agent=lens --input-image=docs/architecture.jpg --json-output