Knowledge limits

The Knowledge page sets how much of a document the Knowledge Bases will index. Each limit protects cost, memory or time. Raise one when a genuinely large document needs to be indexed. You need the admin ability; other people see no access.

Open the page#

1. Select Settings, then the Administration tab. The Knowledge card is shown while Knowledge Bases is on.
2. On the Knowledge card, select Manage.

The limits are in four cards, each with its own Save. Every limit shows what it protects, its Built-in default and the range that is allowed. A limit you have changed shows an Overridden marker. Reset to default puts the default back in the field; select Save to apply it. A number outside the allowed range, or one that is not a whole number where a whole number is needed, is refused with a message in the field.

The limits#

Card	Limit	Built-in default	Allowed up to	What it protects
File limits	Max file size	100 MB	512 MB	Storage and memory. Larger files are refused at upload
File limits	Max extracted characters	20,000,000	1,000,000,000	Cost and search quality. Text beyond it is left out of the document
PDF and images	Max PDF pages	2,000	100,000	Time. Pages beyond it are not read
PDF and images	Max model-read pages per document	300	10,000	Cost. Scanned pages and pictures are read by the image model, and each one is billed
PDF and images	Page images per request	8	100	Memory and request size. How many page images go to the model in one call
Tables	Max rows per sheet	50,000	5,000,000	Memory and time. Rows beyond it on a sheet are left out
Processing	Max decompressed size	512 MB	16 GB	Memory and disk. Office and OpenDocument files are compressed, and this caps how large they may expand
Processing	Extraction timeout	15 minutes	1,440 minutes (24 hours)	Time. Reading a single document stops after this long
Processing	Embedding chunks per request	64	2,048	Request size. How many pieces of text go to the embedding model in one call

The smallest value for every limit is 1. Sizes are entered in MB or GB, the timeout in minutes.

What people see when a limit is reached#

Limit	Result for the person adding the document
Max file size	The upload is stopped as soon as it passes the limit and refused: "This file is larger than the size limit for uploads." A linked file over the limit is left out of a web source, which then shows Partially synced
Max extracted characters	The document is indexed with a note: "Only the first N characters were read because of the file limit set by your administrator."
Max PDF pages	The first N pages are read and the document shows a note: "Only the first N of M pages were read because of the file limit set by your administrator."
Max model-read pages per document	Indexed with a note that names the limit: "Pages 301-420 were not read because they exceed the limit of 300 pages set by your administrator." (pictures read the same way). Pages with a text layer are never counted
Max rows per sheet	Indexed with a note: "Only the first N rows of sheet were read because of the file limit set by your administrator."
Max decompressed size	The file shows Failed: "This file is too large once unpacked"
Extraction timeout	The file shows Failed: "This file took longer than N minutes to read. Try splitting it or asking an administrator to raise the reading time limit." N is the current timeout. It applies to each step of reading a file, including preparing the pages for the image model
Not a limit: a file too heavy for one reading process	The file shows Failed: "This file is too large or complex to be read". The reading process ran out of memory or stopped. Splitting the file usually helps
Not a limit: the image-reading model is unreachable	When every page of a scan fails for the same reason, the file shows Failed with one of: "The model that reads page images could not be reached.", "...took too long to answer.", "...was too busy to answer. Try again later." or "...declined to read this file." The technical detail is in the server log. Check the image-reading model in the AI provider settings.
Page images per request, Embedding chunks per request	Nothing visible. These change how the work is batched

Effect on existing documents#

A limit applies to files that are uploaded or replaced after you save it, and to linked files a web source collects after that. The limits in force when a file is queued are the ones used to read it, so a change never alters a file that is already being read. Documents that are already indexed keep their previous result until someone who manages the Knowledge Base chooses Read again on them, which reads the file again from its original with the limits and the image reading model in force at that moment (see Adding and editing documents). Rebuild search index does not re-read files, so it does not apply new limits. To index a document that an earlier limit cut short, raise the limit and then choose Read again on it, or replace the file. Reading again is billed by the image model like the first read.

The ceilings cannot be exceeded from this page.

Related#

Scanned pages and pictures are read by the model chosen under Image reading model; see Default models. What that reading costs is described in The Usage & cost view.