Incident: A model change the reset never saw
Module 4 · Incident file · About 30 minutes
What happened
AnythingLLM is an open-source application from Mintplex Labs for chatting with your own documents. In December 2024 a user asked how to clear documents after changing the embedding model (issue #2745, 1 December 2024). The answer was pull request #2819, merged on 16 December 2024. Its description says that changing embedding models or vector databases "causes mismatching of vector dimensions", so the change "clears the vector db of all vectors and purges all cached documents". From then on, changing the model on the settings page warned the user, then cleared the documents and vectors.
On 22 September 2026 a contributor reported that one path skipped both (issue #6460). With Azure OpenAI as the embedder, "changing only the embedding deployment name in Embedding Preference skips both the warning and the vector reset." The Azure form saved its deployment under its own field name, and that field wrote the model setting "without the handleVectorStoreReset hook". The report states the consequence: "the old vectors stay next to the new embedder." It dates the gap to v1.4.0, the release of January 2025.
The reporter ran the real settings endpoint on v1.16.2 with one embedded document: after the Azure change, one document and one vector remained; after the generic field, none. They "did not run this against a live Azure resource", and the issue reports no affected users. The fix, pull request #6461, added the hook to the Azure field and was merged on 25 September 2026.