Attributes Agent
Typed product parameters — weight, dimensions, material, power rating — are what makes your feed filterable. Google Shopping, Heureka, Zboží and Allegro all have typed fields for these; a feed without them lands outside the 'notebooks 15"' or 'weight < 1 kg' filters shoppers actually use. The Attributes Agent extracts these specs from descriptions (and, when needed, manufacturer pages) and stores them against a strict canonical vocabulary.
What it looks like in Activity
Example
(no typed attributes)
display_size_inches: 15.6 · ram_gb: 32 · storage_gb: 1024 · product_weight_g: 1860
How it works
Each attribute must use a canonical key from a ~30-key vocabulary (product_weight_g, display_size_inches, material_composition, …). The agent pulls values straight from the description first, then falls back to the manufacturer page if fewer than three attributes were found. Units like kg, lbs, mm and inches get normalized server-side — you never end up with a mix of grams and kilograms in one column.
When it skips
Values that can't be grounded in the description or a fetched source page are skipped, never invented. Range checks (weight 0.1 g – 500 kg) reject obvious parse errors, and the composition validator rejects materials that don't sum to 100%. Missing is always preferred to fabricated.
Typed attributes unlock price comparison filters on Heureka/Zboží/Allegro and Google Shopping's 'shop by' facets — products without them are invisible to the 60-80% of shoppers who filter by size, weight or material. They also feed directly into Google Shopping's g:product_weight, g:product_length, g:material and g:energy_efficiency_class fields.