> ## Documentation Index
> Fetch the complete documentation index at: https://guide.garde.app/llms.txt
> Use this file to discover all available pages before exploring further.
# Data Agents: automatic cleanup
> Background agents that catch duplicates and fill in missing conversions and prices.
**Who:** whoever owns data hygiene — usually the Items owner.
**When:** weekly skim; deeper pass after any big import.
Agents run in the background to detect duplicates and fill missing conversions and default prices — the catalog stays clean without you hunting for problems.
## Where
**Inventory → Reports → Data Agents**. Tabs: **Inventory Items**, **Vendor Items**, **Vendors**, **Estimates**.
## Duplicates
The first three tabs are review queues of likely duplicate groups (the **Inventory Items** tab shows recipe and vendor-item counts to judge by):
* **Merge** — pick the keeper; associated records move to it, the rest are deleted. Cannot be undone.
* **Ignore** — tells the agent the pair is genuinely different; it stops asking.
## Estimates
The **Estimates** tab reviews what the AI filled in:
* **Default Prices** — for items with no purchase history. Saving a manual value drops the row from the AI queue.
* **Unit Conversions** — override with a manual quantity or remove.
* **Vendor Conversions** — open the linked item to enter an exact conversion, or remove.
Every estimate keeps its **history**, an applied estimate can be **reverted**, and prices from unapproved invoices are excluded from estimates — reviewing is never a one-way door.
**Run All Estimates** queues a mass estimate across the chain; a progress banner tracks it while you keep working.
Estimates are scaffolding, not truth. Anything **AI Estimated** feeding a high-volume recipe deserves a real-world check — weigh the case, read the label, overwrite.