Releases: jlueschow/obsidian-euridian-plugin
Release list
0.4.2
Full Changelog: 0.4.1...0.4.2
0.4.1
0.4.0
0.3.0
What's new
Read documents in your vault. The vault agent can now read the text of PDF, Word (.docx) and PowerPoint (.pptx) files (list_documents / read_document), e.g. invoices, reports or slide decks that are not notes. Extracted text is capped like read_note; scanned PDFs without a text layer return no text.
Backend and model are now per chat. Every chat tab keeps its own backend and model. Switching the model in one tab no longer changes it in other tabs or windows, the header dropdown always shows the model that is actually used, and changing the backend in Settings only affects new chats (existing chats keep theirs). Closing the Settings tab refreshes open chats.
Model switches no longer break on images. If a model rejects the images in a chat's history (no vision support, or a per-request image limit such as "At most 3 image(s)"), Euridian now adapts automatically: it retries once without images or with only the newest ones and shows a notice, instead of failing every following message with HTTP 400.
Fixes
- Browser autofill could overwrite the free-text model-name field with an unrelated saved value, which then showed up as a bogus single entry in the chat's model dropdown. Autofill is now disabled on all backend configuration fields.
Notes
- Existing chats are migrated automatically: they keep working with the model that was selected in Settings before the update.
- Build tooling: esbuild updated to 0.28 (needed for
pdfjs-dist), the pdf.js worker is generated at build time.