From the documentation you shared, PyAirbyte is introduced and shown only for basic usage (installing, creating a source, selecting streams, and reading data).[[Using PyAirbyte](https://docs.airbyte.com/developers/using-pyairbyte)] The examples demonstrate:

python import airbyte as ab    source = ab.get_source(     "source-faker",     config={"count": 5_000},     install_if_missing=True, ) source.check() source.select_all_streams() result = source.read()

However, the provided docs and API reference link do not describe any mechanism (argument, method, or option) to: - set a global “max N rows per stream” for a read, or - limit the number of records per table/stream during sync via PyAirbyte.[[Using PyAirbyte](https://docs.airbyte.com/developers/using-pyairbyte)] Because of that, I can’t confidently point to a PyAirbyte-based way (in the documented API) to enforce a fixed row cap per table for a connection. You could always post‑process the returned records in your own code, but that’s outside what the docs explicitly describe. So, based on the available documentation, there is no documented, built‑in PyAirbyte feature to limit rows per table/stream during sync.

Airbyte itself does not provide a generic “sampling” mode for all connectors or connections (e.g., “sync only N random rows per table”). The available information shows: - For connections in general, there’s no built‑in way to limit or sample a small number of records per stream. A similar request (“sample sync connection to return only one record”) was answered with “there’s no straightforward way to do this,” and a feature request was recommended instead.[[Limit sync records](https://discuss.airbyte.io/t/4082)] - Airbyte follows an ELT paradigm, expecting to pull full datasets first and then transform or aggregate in the destination; being “more selective at the source” is not how it’s currently designed, apart from filtering and using views.[[Using dbt transform](https://discuss.airbyte.io/t/3455)] That said, there are a few connector‑specific or pattern‑based workarounds that approximate sampling: 1. Use a view on the source For databases, you can create a view that already restricts or samples the data (e.g.,

LIMIT N

, date filters, or vendor‑specific sampling functions) and then sync that view instead of the full table.[[Using dbt transform](https://discuss.airbyte.io/t/3455)] This is the main officially suggested way to be selective at the source. 2. Use the /dev/null destination to inspect “sample” data If you only need to inspect data and schema (not persist it), the

/dev/null

destination can log first N entries, every N‑th entry, or a random sample percentage per stream, up to a configurable max entry count.[[/dev/null destination](https://docs.airbyte.com/integrations/destinations/dev-null)] - Mode: Logging → “First N entries”, “Every N‑th entry”, or “Random sampling”. - This doesn’t reduce what’s read from the source, but it does let you see a small, sampled subset of records for analysis/debugging. 3. Source‑specific sampling knobs Some connectors have their own sampling‑like config. Example: the Sample Data (Faker) source lets you configure

count

and

records_per_slice

to control how many records are generated per stream per sync.[[Sample Data source](https://docs.airbyte.com/integrations/sources/faker)] This doesn’t help with real production sources, but shows that sampling has to be implemented per‑connector rather than as a platform‑wide feature. 4. Mappings row filters (not true sampling) Airbyte mappings can filter rows based on conditions on string/number fields, mainly for quality/irrelevance/compliance reasons.[[Filter rows](https://docs.airbyte.com/platform/using-airbyte/mappings#filter-rows)] This lets you exclude some data, but it cannot express “take only 10 rows” or “random 1% sample”; it’s predicate‑based, not limit‑based. Given what’s documented: - There is no generic, built‑in sampling mechanism (e.g., “limit N rows per stream” or “sample X%”) for all sources/destinations. - For quickly understanding schema and example records without a full replication, your best options are: - Create source‑side views that expose only a small/sample subset and sync those.[[Using dbt transform](https://discuss.airbyte.io/t/3455)] - Use the /dev/null destination in logging + random sampling / first N modes to inspect sampled records without persisting all data.[[/dev/null destination](https://docs.airbyte.com/integrations/destinations/dev-null)] If you need a platform‑level sampling feature, the documented guidance is to open a feature request (as suggested in the “Limit connection sync records returned” discussion).[[Limit sync records](https://discuss.airbyte.io/t/4082)]