GitHub Repo
Langflow 1.13.2 Fixes Duplicate Knowledge Base Imports With Stable Document IDs to Protect Retry Data
Released on November 10, Langflow 1.13.2 reuses stable document IDs when files and memory data are imported repeatedly, protecting completed chunks. The update does not remove existing duplicates, and outdated chunks from modified content still require separate handling.

Langflow released version 1.13.2 on November 10, including a fix to document identity in Knowledge Base imports. For workflows that repeatedly upload files, periodically add agent memory, or retry after failures, the new version uses stable document IDs so identical chunks can be recognized and duplicate writes avoided. The fix was merged on November 4 and is now available to users in the official release. Release announcement
The fix centers on checking state before writing. According to the maintainers, the import path checks existing documents and the status of their associated jobs, preserves metadata for completed chunks, resumes chunks from failed jobs, and rejects overlapping imports that are still in progress. This prevents a failed retry from deleting previously stored data. The system also counts “processed” and “newly written” chunks separately, so partial-write checks and new-item counts use the appropriate measures; table imports retain their existing duplicate-data setting. Fix details
For retrieval-augmented generation (RAG), this makes data updates easier to verify. The official documentation says the Knowledge Base component provides import and semantic search modes, can identify rows through the Identifier field, and includes a setting that allows duplicate data. From an engineering perspective, repeated retries that keep adding the same content could affect index size and retrieval candidates, so document identity and new-item counts are worth including in import monitoring. Component documentation
The upgrade still has clear limitations: existing random IDs and duplicate data are not rewritten, and the first re-import of old data may create an additional copy with a stable ID. Changes to content also do not automatically remove outdated chunks. The maintainers reported that three query regression tests went from failing to passing, but broader testing still had failures reproducible on an unmodified branch, as well as configuration errors; validation against actual PostgreSQL and OpenSearch deployments was also listed as pending. Engineering teams should check retry behavior, chunk counts, and old-data cleanup strategies for their own backends, and should not treat this update as a complete document synchronization mechanism. Validation and limitations