1.17.1

Fixes

• fix: clarify empty settings states and drop the empty frontmatter group

1.17.0

What changed

• Requires Obsidian 1.14.4 or newer. Older versions stay on 1.16.1.
• Settings rebuilt on Obsidian's native groups. Providers, Decision API, Chat and Frontmatters now line up the same way and follow your theme. Each group has a + in its header, and adding from Decision API starts with the decision toggle on.
• Hidden notices. The row only appears while you have muted notices, and Show again restores them.

Commits

Fixes

• fix: start the decision toggle from the group a model is added in
• fix: keep the divider between tags and custom frontmatter rows
• fix: right-align the tags edit button and show hidden notices only when any exist

Other

• refactor: use native groups for frontmatter rows instead of overrides
• refactor: build settings on SettingGroup and require Obsidian 1.14.4

1.16.1

Fixes

• fix: align settings rows and show which models are in use

1.16.0

What changed

• Model insights is gone. The chart, price catalog and Refresh prices are removed, so the plugin no longer contacts models.dev.
• Usage on provider rows. Each provider now shows tokens and requests from this plugin over the last 30 days, plus recent response time. Counts start from this version and cover this plugin's requests only. Note text is never stored.
• Clearer settings. Models are grouped into Decision API and Chat, and each frontmatter field is its own card.

Commits

• feat: group models by role and separate frontmatter rows
• feat: replace model insights with usage on provider rows

1.15.1

Fixes

• fix: stop the default rule from emptying generated tags
• fix: send requests current OpenAI and Anthropic models accept

1.15.0

Features

• feat: add model insights with pricing and a comparison chart
• feat: replace retired preset models in saved settings
• feat: keep stable prompt text first so providers can cache it

Fixes

• fix: keep modal focus off refresh and give chart labels room
• fix: keep newer bundled prices and stop chart hover errors
• fix: stop settings re-rendering forever when saved prices exist

Other

• chore: override minimatch to clear the dependency advisory
• refactor: reduce model insights to one chart and one table
• chore: open a weekly pull request when model catalog values change
• chore: refresh model presets to current provider lineups

1.14.0

Features

• feat: classify with decision APIs across providers

Other

• refactor: keep chat models out of decision classification

1.13.2

Fixes

• fix: polish settings modals and draft frontmatter handling

1.13.1

Fixes

• fix: keep the mute when unmuting fails to save
• fix: show focus and pressed states on notice actions
• fix: confirm and report failures when undoing a muted notice
• fix: add inline Undo to the muted notice

Other

• chore: sync vitest eslint plugin with boiler-template
• chore: sync infra files with boiler-template

1.13.0

Classification now runs through Jev, with clearer messages when something needs your attention. Your existing settings and sign-in are kept.

New

• Jev classification. Jev chooses which of your reference values fit a note, for every frontmatter field. Add your Jev API key and model in settings.
• Sign in with your ChatGPT account. Connect through a one-time device code. No browser redirect is needed.
• Custom providers. You can name your own OpenAI-compatible provider, such as a local model, and edit or re-save it later.
• Clear problem notices. Each problem now has its own message, and setup problems have an Open settings button:
  • Jev API key is missing.
  • Jev model is missing.
  • Jev rejected the API key.
  • Jev could not be reached, or is unavailable.
  • The note is too large to classify (the limit is about 24 KB).
  • Settings could not be saved. If sign-in stops working, connect your account again.
• New tags stay optional. New tags are generated by a separate provider, and only when you turn that on for the tags field.

Changed

• Safer note updates. A note is updated in a single step, and nothing is written if the note or your settings changed while classifying.
• Folder-based candidates. When a field has no reference values, the note's own folder supplies them for that run. Your settings are not changed.
• Signed-in status. A Codex account whose sign-in can refresh now shows as connected, with "refreshes on next use".
• Cleaner settings. Duplicate headings are gone, and the provider list explains that it is only used for new tags.
• Tag values. Tag candidates no longer include the leading #.
• Whitespace. Your Jev key and model are trimmed when saved.

Fixed

• Sign-in refresh. If saving settings fails right after a refresh, the new sign-in is kept and the save is retried, instead of going back to an expired one.
• Parallel notes. Classifying several notes at once no longer fails the second one while the sign-in refreshes.
• Older settings. Sign-ins saved by earlier versions are moved to the Codex provider and keep working.
• Shared defaults. Newly added fields no longer share their count and reference lists with each other.

Known gaps

• Long notes are not split up. A note over about 24 KB is reported as too large and is not classified.
• Not tested with live Jev or OpenAI accounts, or with a real device-code sign-in.
• Not tested on mobile, at phone width, or in popout windows.