Product sources

The sources table

Brand → (your brand) → Products → Sources lists every catalogue feeding this brand.

Column	What it tells you
Name	The source name, with its address underneath. A push-only API source shows its push address instead.
Type	CSV feed or Product API.
Status	See the badges below.
Refresh	"Every day", "Every 3 days", "Push only", or "Stopped after 3 failures".
Last successful fetch or push	The date of the last successful read. A push-only source shows "Pushed …" with the date your shop last sent us changes, instead of "Never".
Products	How many base products this source currently holds. Versions are not counted here.

Status badges

Badge	Meaning
PENDING	Created, but nothing has been read from it yet. A push-only source leaves this state as soon as your shop's first push succeeds.
PROCESSING	A read or a push is running right now.
ACTIVE	The last read succeeded.
WARNING	The last attempt failed. We keep retrying every hour.
INACTIVE	Stopped after three failures in a row.

Hover the information icon beside a badge for the plain-English reason, including the raw error when there is one.

While anything is PROCESSING, the table refreshes itself every few seconds — press Fetch now and watch the row finish without reloading the page.

Adding a source

Press Add source. The first choice is the source type, and it is the one thing you cannot change later.

Fields for every source

Name

required

What you call this source. Up to 200 characters.

Address

required

Where we read your products from. A Product API address must start with https://; a CSV feed may use plain http. Leave it empty on a Product API source to make it push-only: the schedule switch turns itself off and cannot be switched on, no first read is started, and the source simply waits for your shop to send it products.

Read every … days

Between 1 and 30. The default is every day.

If the source keeps failing

What happens to your products once reading has stopped. See below.

Read it straight away

On by default. Starts the first read as soon as you save.

Extra fields for a Product API source

Shared secret

Between 16 and 256 characters. Leave it empty and we generate one for you. We send a SHA-256 hash of this secret in a Signature header on every read, so your shop can check the request really came from Yaplet. The secret itself never travels.

Extra headers

Up to 20 name/value pairs sent with every read — an API token your shop expects, for example. You cannot override Signature, Host or Content-Length.

Fields that become keys

The product fields whose values should become the product's keys — an ingredients field, for instance. See below.

Scheduled reading

Switch it to Push only and we never read on our own; your shop sends us changes instead.

Fields that become keys

This is the setting that connects your catalogue to your product rules.

A rule applies to a product when the two share a key. A product gets its keys from two places: the rule_keys list your shop sends with it, and the values of any product fields you name here.

So if your shop sends an Ingredients field and you add Ingredients to this box, every ingredient becomes a key on that product — and any rule carrying that ingredient's name applies to it automatically.

After a source has run once, the form shows a row of chips listing every field name your shop actually sent. Click one to use it, click it again to drop it — no typing and no guessing at spelling.

Changing this setting re-derives the keys of every product of that source immediately. There is no refresh to wait for and nothing is re-read.

If the source keeps failing

This choice only takes effect once reading has stopped after three failures in a row.

Keep serving the last products we read

default

Your AI keeps recommending what it already knows, which may have gone out of date.

Hide the products until it works again

The products disappear from your AI and from product cards, and come back on the next successful read or push.

Products you switched off by hand are never switched back on by this.

What happens when a read fails

Within a single read we try three times — 30 seconds each, waiting 5 then 15 seconds between attempts — but only when the problem looks temporary (a timeout, a network error, or a server error on your side). A rejection like "not found" or "forbidden" fails at once.

Then:

	What happens
1st and 2nd failure in a row	Status WARNING. We keep retrying every hour. Nobody is told.
3rd failure in a row	Status INACTIVE. Scheduled reading stops. Your organisation's owner gets a notification and an email.

A read that comes back with no products at all counts as a failure, not as "delete everything". A broken feed can never silently empty your catalogue.

Two things deliberately do not count as strikes: a Fetch now you ran yourself (it only sets WARNING, and nobody is emailed), and a failed push from your shop (it shows in the run report only).

The email you get

One email per organisation per hour, however many sources are involved — it cannot nag you. It tells you what happened, why, how to fix it, and what happens if you do nothing, with a button straight to this page.

A successful push does not restart a schedule that has stopped. Once a source is INACTIVE, only Fetch now starts it reading again.

Fetch now

The circular-arrow button reads the source immediately, in the background.

There is a five-minute cooldown per source. Inside it, the button tells you how long is left.
It is also how you restart a stopped source: a successful manual read clears the strikes, sets the source back to ACTIVE, and brings back any products that were hidden.

The last run report

The chart icon opens the report for the most recent run. Only the latest run is kept for each source — read it before you trigger the next one.

It shows whether the run was scheduled, manual or a push; whether it was a full snapshot or only changes; and a grid of counters.

The ones worth knowing:

Counter	What it means
Items received / Items used	How many products arrived, and how many survived validation.
Products added / updated / unchanged / removed	What changed at the base-product level.
Versions stored / Versions removed	The same, for versions. On a healthy run this is at least as large as the product count.
Search card pieces indexed	How much searchable text the AI re-read. A product becomes one to a few pieces depending on how much it carries.
Overview pages rewritten	The catalogue pages that were rebuilt — one for the catalogue as a whole, plus one per category.
Skipped as invalid	Items that were left out. The reasons are listed underneath.
Versions that stopped a product	A version that failed validation takes its whole product out of the run, so this being non-zero explains a missing product.
Product prices ignored (versions sent)	You sent a price on the product and versions underneath it. The versions win and the product-level price is ignored — see the Product API reference.
Search terms cut for length	A product sent more search terms than one product may carry; the rest were dropped.

If the products saved but the catalogue pages did not, the report shows: "The products were saved, but the catalogue overview pages could not be updated. The next run rebuilds them." Your products are fine and answers still work — but until the next run, the AI's picture of your catalogue's structure is one run out of date.

Below that, up to three lists:

Skipped items — which items were left out and why (the first 50).
Products cut for length — products that sent more keys or more search terms than one product may carry.
Fields the shop sent — every field name seen in this run. This is the same list the "fields that become keys" chips come from.

Row actions

Fetch now, Last run report, Edit, Delete — plus two more on Product API sources: Copy push URL, and a shortcut to the API tab with that source already selected, so the code samples carry its real address.

Deleting a source deletes its products, along with everything the AI learned about them. The confirmation tells you exactly how many. A source cannot be deleted while a read or push is running.

Limits worth knowing

Limit	Value	What you see if you hit it
Products per source	20,000	The read fails and asks you to split the catalogue across several sources
Products per organisation	20,000	The read fails; write to support to have it raised
Versions per organisation	100,000	The read fails, naming the number it would have reached
Versions per product	500	The product is skipped as invalid
Sources per organisation	100	Saving the source fails
Response size	50 MB	"The response is larger than 50 MB"
Read timeout	30 seconds per attempt	"The request timed out after 30 seconds"
Extra headers	20 per source	The Add header button stops
Fields that become keys	20 per source	The box stops accepting more
Keys per product	500	Counted as "Keys cut for length" in the report
Search terms per product	100 words, versions included	Counted as "Search terms cut for length" in the report
Category depth	6 levels, 100 characters each	The item is skipped as invalid
Description per product	20,000 characters	The item is skipped as invalid

There is no cap on custom field text any more. Your product's own fields used to be cut at 4,000 characters; now everything you send is kept and indexed, split into search pieces behind the scenes. The old "Fields cut for length" counter only appears on reports from before this change.

Smaller limits while nobody is paying

An organisation with no active subscription and no credit balance gets deliberately smaller feed limits, because it could never store a catalogue that size anyway:

Limit	While nobody is billed	On a paid plan or with credits
Response or push size	5 MB	50 MB read / 25 MB push
Products in one read or push	1,000	20,000
Pushes per minute	6	60

Adding credits or starting a subscription lifts these immediately — there is nothing to switch on and nothing to wait for.

The API tab prints the numbers that apply to your organisation, not the paid ones — so a developer reading it on a free organisation sees the real limits. When the smaller set is in force, the tab says so and names what the limits become once you add credits.