Document Quota
The document quota caps the total number of documents across the workspace’s knowledge bases, whether you upload them or import them from a source. Every document counts, including ones that are disabled, archived, or failed to process. At the quota, uploads and imports that would add new documents fail. To add more, delete documents you no longer need or upgrade your plan.Knowledge Data Storage
Each document is split into chunks, the passages a search returns. Knowledge data storage holds those chunks in searchable form: their text and their vectors. A vector is the numeric form of a chunk that semantic search matches by meaning. Your original files don’t count, so usage depends on how much your documents produce once indexed, not on their file size.What Uses Storage
In a classic knowledge base, documents indexed with the High Quality index method store a vector for each chunk, or for each child chunk in Parent-child mode. Documents indexed with the Economical method use keyword indexing instead, which doesn’t count toward knowledge data storage. In an agent knowledge base, every chunk is stored both as a vector, for matching by meaning, and as text, for matching by keywords. The knowledge graph adds vectors for the subjects it maps and how they connect. In both kinds, a vision-capable embedding model adds a vector for each image. The embedding model also sets how large each vector is: OpenAI’stext-embedding-3-large, for example, by default produces vectors twice the size of text-embedding-3-small’s.
The usage figure isn’t live and can take several minutes to catch up after an upload or deletion.
When a Classic Knowledge Base Upload Would Exceed the Limit
Before indexing, Dify may refuse an upload based on how much storage it’s estimated to take. The refused document stays in the list with an Error status. To index it, upgrade your plan or free up storage, then click Retry above the document list. Space you free up may not count toward a retry right away, especially while other documents in the workspace are still being indexed. You can also delete the document and upload it again with a larger chunk size. That can reduce the number of chunks, but larger chunks can also make each retrieved passage longer and less precise.When Storage Is Full
Once usage reaches your plan’s limit, neither kind of knowledge base can add content:- Classic knowledge bases block uploading documents, adding or editing chunks, and re-indexing, such as after changing chunk settings. In Dify, enabling, disabling, archiving, and restoring documents are blocked too, though the Service API can still change their status. This holds even in knowledge bases that use Economical indexing, because the block covers the whole workspace.
- Agent knowledge bases can’t process new or changed content. New uploads and imports may be refused outright, and documents already processing, syncing, or re-indexing fail. An embedding model change can’t run either. Searching and deleting keep working. Once there’s space, add refused content again and retry the documents that failed.