JMN Field Manual

01.2 — Getting Started

Creating Your First Agent

Six steps from an empty workspace to your first AI-generated answer, grounded in a real document. Follow this once and the rest of the manual will make sense by reference.

  1. Create the agent Click New agent in the sidebar, give it a name and a short description of what it should extract, classify, compare, or validate. The description is just for humans browsing the catalogue later — it isn't sent to the AI.
    2.1
    The Create new agent dialog

    FIG. 2.1 — The "Create new agent" dialog.

  2. Add an input property This is your first column, and it should be a file. Click Add property, set its direction to Input, and name it after what it holds — "Contract PDF," "Invoice," "Loan Agreement."
    2.2
    An empty agent grid right after creation

    FIG. 2.2 — A fresh agent, no properties yet.

  3. Upload a file Add a row, then drag a PDF onto the input cell. It ingests in the background — extracted, chunked, and embedded — and the cell's status badge moves from queued to processing to done.
    2.3
    A file mid-upload showing an ingestion status badge

    FIG. 2.3 — Uploading and ingesting a file.

  4. Add an output property Add a second property, set its direction to Output, choose an output type (start with Text), and write a prompt that references your input column by name — e.g. @Contract PDF — what is the termination notice period?
    2.4
    Adding an output property with a prompt box

    FIG. 2.4 — Writing the first prompt.

  5. Run the cell Click the cell to run it. While it's working the cell shows a generating state; this happens in the background, so you can keep working elsewhere in the grid.
  6. Read the result The finished cell shows the answer with small numbered citation markers. Click one to jump straight to the exact passage in the source PDF that the answer came from.
    2.5
    A finished output cell with citation markers

    FIG. 2.5 — A grounded result, ready to click into.

Tip

Add a few more rows and the prompt you just wrote runs against each of them with no extra setup. This is the entire point of building it as a column instead of asking once in a chat window.