Skip to content

ObsidiAI 0.1.6

Choose a tag to compare

View all tags
@github-actions github-actions released this 20 Sep 07:24
· 6 commits to main since this release
0.1.6
f19a915

ObsidiAI 0.1.6

Attach files, drop files, paste images

  • Attach files from computer selects multiple UTF-8 text, Markdown, CSV/TSV, JSON/YAML/TOML, log, source/config files, or PNG/JPEG/WebP/GIF images.
  • Drag and drop files onto the chat to use the same attachment flow. Clipboard images can be pasted directly into the composer.
  • Images reach the agent as real image inputs. Image-containing conversations require a model advertising image support; text-only models are rejected before consuming the draft rather than silently dropping images.
  • Click attachment chips to preview safe raw text or images, before sending or after reopening saved history. A separate remove button removes draft attachments. Attachment-only messages are supported.

Compact pasted text and native navigation

  • Pasting 4,000 characters or 40 lines creates a temporary .txt attachment with a clickable preview instead of expanding the message. Short pastes stay normal text; text outside the selected draft range is preserved.
  • Temporary attachments are in-memory snapshots, not new vault notes or operating-system temporary files.
  • Ctrl-click / Cmd-click internal note links and embedded notes in assistant output to open a new tab without replacing the chat. Generated tool-result note links also support this behavior.

Limits and privacy

  • Maximum 200,000 characters per text attachment, 5 MiB per image, 20 attachments, and 20 MiB total attachment data per message. Files are not silently truncated. Unsupported inputs and failed reads are reported; late reads after view closure are discarded.
  • Text decoding rejects malformed UTF-8 and binary/control-byte content. Recognized source files tolerate incorrect platform MIME labels; image signatures must match their advertised format.
  • Nothing is uploaded merely by selecting, dropping, or pasting. External attachments retain only the filename, not the computer directory. On Send, content is sent to the selected provider and saved in the plugin's unencrypted history, which vault sync and backups may copy.
  • Sent previews reference existing model-message blocks rather than duplicating large text/image payloads in timeline storage. Existing saved history remains readable; unsent attachments are not persisted.

Install or update

  1. Download obsidiai-0.1.6.zip, not GitHub's automatic source archive.
  2. Replace main.js, manifest.json, and styles.css in <vault>/.obsidian/plugins/obsidiai/. Preserve data.json and history.json if present.
  3. Reload Obsidian or disable/re-enable ObsidiAI after the current run settles.

Requirements remain Obsidian desktop 1.11.4+ and embedded Node 22.19.0+. The three individual plugin files are also attached.

Verification

  • 140 tests cover the integrated implementation, including multimodal transport, attachment-only submission, preview/history roundtrips, rejected image models, attachment bounds, invalid text/image inputs, and CRLF paste thresholds.
  • Native Obsidian 1.13.7 checks exercised the actual file picker, clipboard image/text paste, file drag-and-drop, safe previews, compact sent messages, history reload before model setup, failed-input draft retention, late-read disposal, and Ctrl/Cmd-click note and embed navigation.
  • Light/dark previews and narrow layouts were visually checked. Native transport checks used a local scripted provider; no remote vision-provider or new OAuth verification is claimed. Actual third-party plugin lifecycle effects remain unverified.
Assets 6
Loading