Run the agents on a sample panel

Three agents read the inputs, propose a layout and write the review. Each one is a real task run through HarnessRouter while you watch. Code checks every answer before the next step starts.

The inputs

Pick a sample. These are generic panels made for this demo, each with a deliberate problem for the agents to find.

P&IDGeneric sample
Bill of materials
ItemDescriptionQty

The reader agent gets these two files: pid_tags.csv (the tags and their positions, read off the drawing by code) and bom.csv.

The sample design rules

    The run

    Checking whether live runs are available…

    • Agent task on HarnessRouter
    • Plain code
    • You, as the engineer

      What is real on this page

      The agent steps are real. Read, lay out and review each start a task on HarnessRouter and wait for it to finish. The response ids in the run record match the task list in our HarnessRouter workspace.

      Each step can use a different model. The server maps a step to an agent setup: which harness, which model, how much reasoning. Changing a model is a one-line change, with no change to the pipeline.

      The checks are code. Whether the tags were read correctly, whether the layout follows the rules and how many parts it needs are computed, not asked of a model. A wrong answer stops the run and is shown.

      The panels are samples. In the product, the inputs are a customer's own P&ID, BOM, parts library and rules, and the build step creates a SolidWorks assembly on a licensed Windows machine. Here it stops at a 3D preview with generic parts.