What Mintplex-Labs/anything-llm shipped
Written by FoxPlug from the repository's own releases; not affiliated with Mintplex Labs. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/Mintplex-Labs/anything-llm; Mintplex Labs did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of October 5, 2026
What shipped
- Scheduled jobs now stop when an instance switches to multi-user mode, instead of leaving their cron timers registered and starting runs that the admin could no longer reach. Pull request #6656
- Document embedding for AstraDB works again after the collection check was moved back onto the instance. Pull request #6654
- Confluence load failures now surface the real error instead of reporting that no pages were found for the space. Pull request #6652
- Bulk scraped pages that share a file name are written to separate files rather than overwriting each other. Pull request #6650
- Chat attachment state and uploads are now isolated per conversation. Pull request #6635
- The agent rag-memory search now applies the workspace Document similarity threshold instead of always filtering at the default. Pull request #6648
- Scheduled job responses no longer include raw think blocks after tool calls. Pull request #6663
- The document picker loads the next page from the server's offset so Load more no longer repeats or skips files. Pull request #6628
- Requested files are now attached to Outlook replies and reply drafts. Pull request #6623
- An unknown workspace on the manage-users and update-pin routes now returns 404 instead of a 500. Pull request #6625
Why it matters
Most of this week is fixes rather than new features, and several of them affect day to day use: scheduled jobs, document embedding, Confluence loading, bulk scraping and chat attachments. The changes remove cases where a failure looked like success, such as Confluence reporting no pages or a scrape reporting a page it had overwritten.
Changelog entry
- Stop scheduled jobs after switching to multi-user mode Pull request #6656
- Restore document embedding for AstraDB Pull request #6654
- Surface Confluence load failures instead of reporting no pages Pull request #6652
- Write bulk scraped pages that share a file name to separate files Pull request #6650
- Scrape each page once when bulk links differ only by a fragment Pull request #6619
- Isolate chat attachment state and uploads by conversation Pull request #6635
- Preserve referenced MBOX chat attachments during orphan cleanup Pull request #6633
- Apply the workspace similarity threshold to agent rag-memory search Pull request #6648
- Finalize stopped scheduled job runs on Windows Pull request #6637
- Fix raw think blocks in scheduled job responses after tool calls Pull request #6663
- Load the next page of documents from the server's offset in the picker Pull request #6628
- Attach requested files to Outlook replies and reply drafts Pull request #6623
This week: scheduled jobs stop on multi-user switch, AstraDB embedding works again, Confluence errors show up, bulk scrapes stop overwriting each other, chat attachments isolated per conversation, rag-memory uses your similarity threshold.
This week brought a set of fixes across scheduled jobs, connectors, scraping and chat. Scheduled jobs now stop when an instance switches to multi-user mode, instead of leaving timers registered and runs unreachable. Document embedding for AstraDB works again. Confluence load failures now surface the real error rather than reporting no pages. Bulk scraped pages that share a file name are written to separate files instead of overwriting each other. Chat attachment state and uploads are isolated per conversation. The agent rag-memory search applies the workspace similarity threshold. The document picker loads the next page from the server offset. Requested files are attached to Outlook replies, and an unknown workspace on the admin routes returns 404 instead of 500.
Week of September 28, 2026
What shipped
- Added reasoning effort controls for configurable inference behavior. Pull request #6269
- Workspace temperature setting now applies to agents and omits the parameter when unset. Pull request #6285
- Empty chat greetings now rotate based on time of day instead of showing a static message. Pull request #6567
- Fixed uploaded images persisting when switching from chat to agent mode mid-session. Pull request #6572
- Scheduled jobs now correctly shift weekday and day of month across UTC midnight boundaries. Pull request #6558
- Hourly schedule minutes now convert to UTC in half-hour time zones like Asia/Kolkata. Pull request #6605
- API document uploads now embed every document returned by the collector, including all spreadsheet sheets and mbox messages. Pull request #6532
- History attachments no longer sent to providers that build prompts from raw chat history. Pull request #6591
- Kubernetes probes now point to the server's /api/ping endpoint instead of the collector. Pull request #6592
Why it matters
This week focused on reliability fixes across scheduling, document handling, and agent functionality. Multiple fixes address edge cases in timezone handling, file uploads, and integration points that improve the stability of production deployments. The v1.17.0 release consolidates these improvements with new reasoning and temperature controls.
Changelog entry
- Release v1.17.0 Release
- Add reasoning effort controls Pull request #6269
- Apply workspace temperature to agents and omit when unset Pull request #6285
- Rotate empty chat greetings by time of day Pull request #6567
- Fix scheduled jobs shifting weekday and day of month across UTC midnight Pull request #6558
- Fix hourly schedule minute conversion to UTC in half-hour time zones Pull request #6605
- Keep uploaded images when switching from chat to agent mid-session Pull request #6572
- Embed every document from API uploads into addToWorkspaces Pull request #6532
- Point Kubernetes probes at server's /api/ping endpoint Pull request #6592
- Stop sending history attachments to providers building from raw chat history Pull request #6591
- Save turning password protection off so it survives restart Pull request #6562
- Render C++, ~~~ and nested fenced code as code in Telegram replies Pull request #6547
V1.17.0 ships with fixes for scheduled jobs, document uploads, agent attachments, and timezone handling. Better support for reasoning controls and workspace temperature settings.
AnythingLLM v1.17.0 is out. This release addresses critical issues in scheduled job execution across timezones, agent attachment persistence, and API document uploads. New features include reasoning effort controls and workspace temperature applied to agents. Dozens of edge case fixes improve reliability for document handling, Telegram integration, and embedded deployments.
Week of September 21, 2026
What shipped
- AnythingLLM v1.16.2 released with LLMman OSS as a local provider, Google Vertex AI support, image generation agent skill, and You.com as the default web search provider. Release
- Firecrawl added as an agent web-search provider. Pull request #6480
- AnySearch added as an agent web-search provider for real-time web search. Pull request #6439
- Provider search menus now support keyboard navigation and autofocus. Pull request #6483
- Workspace settings now autosave when edited. Pull request #6343
- Generated images are now included in chat exports. Pull request #6482
- TogetherAI provider now allows configuring
max_tokensto prevent response truncation. Pull request #6463 - Multi-sheet XLSX uploads now preserve all sheets instead of losing non-ASCII sheet names. Pull request #6450
- Weaviate and AstraDB vector scoring now uses cosine similarity consistently with other providers. Pull request #6488
Why it matters
V1.16.2 brings new LLM provider options and web search capabilities, while fixes address data loss in uploads, vector scoring consistency, and document re-syncing. These improvements expand provider choices and fix critical bugs affecting data integrity.
Changelog entry
- Added LLMman OSS as local LLM provider Release
- Added Google Vertex AI as LLM provider Release
- Added generate-image agent skill Release
- Set You.com as default web search provider Release
- Added Firecrawl as agent web-search provider Pull request #6480
- Added AnySearch as agent web-search provider Pull request #6439
- Added keyboard navigation and autofocus to provider search menus Pull request #6483
- Workspace settings now autosave on edit Pull request #6343
- Chat exports now include generated images Pull request #6482
- Fixed multi-sheet XLSX uploads losing non-ASCII sheet names Pull request #6450
- Fixed Weaviate and AstraDB vector scoring to use cosine similarity Pull request #6488
- Fixed text uploads being read incorrectly when using non-UTF8 encoding Pull request #6486
AnythingLLM v1.16.2 ships with new LLM and web search providers, autosaving workspace settings, image generation skills, and fixes for multi-sheet uploads and vector scoring.
AnythingLLM v1.16.2 is out. New features include LLMman OSS and Google Vertex AI as LLM providers, Firecrawl and AnySearch for web search, image generation agent skills, autosaving workspace settings, and exported chat images. Fixes address data loss in multi-sheet uploads, vector scoring consistency across providers, document re-syncing, and encoding issues.