What agents see | Etals | Documentation

An agent can only write a good Description if it knows what the product is. Etals hands the agent the item’s data for you. You don’t need to put variables in your prompts to make that happen.

The short version

Once a content type’s setup is finished, every time an agent works on one of its items, Etals adds a short block about that item to what the agent reads. It holds:

the item’s Name;
every property with Agents can see switched on, written as “Property: value”;
for choice properties, only the options the item has, not the whole list;
the item’s Images, as image files, if images are switched on.

Etals leaves out:

the property the agent is writing right now (so it doesn’t copy the old value);
properties with no value for this item;
properties with Agents can see switched off.

For properties that vary, the agent gets the value for the language, market or website it’s working on.

None of this is sent until the content type’s setup is finished. Go to Setup > Onboarding, open the content type’s chapter, click Set up and finish the wizard with Finish setup. See Content types.

An example

Nordvik runs a task that writes the Swedish Description of the Fjord Shell Jacket (NV-1001). Description is the property being written, so it’s left out. Meta title and Meta description are empty for this product, so they’re left out too. Nordvik has switched Agents can see off for Price SEK, so it’s missing as well. Supplier Notes is included because Nordvik wants agents to use them.

Simplified, the agent receives something like this:

WHAT YOU KNOW ABOUT THIS RECORD:
Name: Fjord Shell Jacket
Category: Jackets
Colour: Deep teal
Material: 3-layer recycled polyester
Gender: Women
Supplier Notes: 3L waterproof shell, 20k/20k, fully taped seams, helmet compatible hood, 2 chest pockets, pit zips. Recycled poly face. PFC free DWR. Weight 410 g (size M).
Image: [file: NV-1001.png]

The real block uses the property names from your workspace and can differ in format. Etals sends it along with the agent’s own prompt, so your prompt can say “Write a product description” and nothing more.

What else the agent gets

The item block is one part of what an agent reads. The others:

Part	Where it comes from	Nordvik example
Field instructions	The prompt of the property being written, from My Data > Properties & prompts or Setup > Prompts > My Data. It’s sent as a separate block headed “FIELD INSTRUCTIONS FOR” and the property name.	”Description: 80 to 120 words, start with the use case, end with care advice.”
Skills	The skills attached to the agent. See Skills.	Brand guidelines, Sustainability claims
Audiences	The audiences picked for the grid, the column or the canvas. See Audiences & tags.	Weekend hikers
The agent’s own prompt	The agent’s setup. See Agents.	”You write product copy for Nordvik Outdoor.”
Item brief	Only for items you create in a task’s grid: what you wrote under “What is it about?” and the files you added. Text from documents and the images go along.	A short brief for a new rain jacket, with the supplier’s spec sheet

Control what agents see

You decide what goes into the item block in two places. Both change the same setting.

In the setup wizard

The wizard’s Properties step asks “Choose what Etals generates and what agents see”. It lists every property with two switches. Turn off Agents can see for what would only distract an agent. The rest of the step:

Setting	What it does
Images	Under More context for agents: “Used to describe colour and fit”. When on, the item’s images are sent to the agent.
Primary language	Below the list. Agents read values in the language they write in first. Where a property has no localization in that language, they read the primary language instead.

More context for agents also has Variants and sizes and Stock switches. They are saved with the setup, but they don’t change what writing agents receive yet. Primary language is used today when the chat assistant reads your data, not yet in tasks.

In the Agents can see column

On My Data > Properties & prompts, the Agents can see column has a check box for each property. A check means agents read it. Click the box to switch it, or use Hide from agents and Show to agents in the row menu. Changes apply to the next agent run.

The Properties & prompts tab for Products with the Agents can see column checked for every property, and the row menu for Price SEK open with Hide from agents

Two switches, two jobs

People mix these up, so here they are side by side.

	Etals can generate	Agents can see
Question it answers	May Etals write this property?	Does an agent read this property while writing something else?
Nordvik example: Supplier Notes	Off. The supplier owns them, and Etals should never rewrite them.	On. They’re great raw material for a Description.
Nordvik example: Description	On. Etals writes it.	On. When the agent writes the Meta description, it can read the Description.

Choosing what to include

Include facts an agent needs to write well: material, colour, fit, use case, supplier notes.
Exclude internal codes, numbers without meaning to a reader, and fields that are often wrong. An agent will repeat bad data.
Turn Images on for products where looks matter. The agent can then describe colour and shape from the photo.
Fill in missing values before you run a large task. Empty values are dropped, so the agent has less to go on.

Variables in prompts

You can still pull a specific value into a prompt yourself. In the prompt editor’s Advanced mode, the Variables panel adds values such as the language, the market or one property of the item. Most prompts don’t need this any more. See Prompts.