ReaderFactory builds a fresh reader on every call

Agno's ReaderFactory.create_reader() now constructs a new reader on every call and hands it the keyword arguments you pass. Earlier releases kept one cached reader per type, so the first call's options won for the rest of the process.

Knowledge picks readers by file extension through this factory, which meant two knowledge bases in one process shared the same reader object. Changing chunk_size on one quietly changed it for the other.

from agno.knowledge.reader.reader_factory import ReaderFactory
 
a = ReaderFactory.create_reader("text", chunk_size=200)
b = ReaderFactory.create_reader("text", chunk_size=1000)
print(a.chunk_size, b.chunk_size, a is b)
 
c = ReaderFactory.create_reader("text")
c.chunk_size = 50
print(ReaderFactory.create_reader("text").chunk_size)

Real output, earlier release and this release:

# Before
200 200 True
50
 
# Now
200 1000 False
5000

ReaderFactory.clear_cache() is now a no-op kept for backward compatibility.

Learn more about reader auto-selection in the documentation.