This repository was archived by the owner on Jul 18, 2026. It is now read-only.
AVE 1.0.14
AVE 1.0.14 Release Notes
Status: Released.
Base: changes after 906550d (Mark AVE 1.0.13 released) through 43f4c0f (Polish local AI model migration).
Highlights
- Added Ask-native graphics actions so safe one-step graphics and design edits can run directly from Ask mode instead of being forced through Plan mode.
- Added guided required-field cards when AVE needs graphic text, titles, handles, or template fields before applying an edit.
- Improved AI graphics updates so selected or referenced graphics can be edited with clearer tool contracts and fewer raw JSON or limbo responses.
- Improved generated sequence caption behavior so caption requests use the full caption-generation path instead of leaving most of a sequence uncaptained.
- Replaced the retired Qwen3.5 2B OptiQ local timeline model with Qwen3 4B Instruct as the default local tool-caller candidate.
- Added Settings cleanup for retired Qwen3.5 2B OptiQ language model files, with Reveal and Delete controls, without re-adding the retired model to normal model choices.
Verification Plan
npm run buildnpm --prefix render-worker run buildnode --check electron/main.cjsnode --check electron/preload.cjsnode --check scripts/assert-notarization-env.cjsnode --check scripts/audit-desktop-package.cjsnpm run ai:prompt-smokenpm run ai:prompt-regressionnpm run test:runnpm audit --jsonnpm --prefix render-worker audit --jsongit --no-pager diff --checkenv APPLE_KEYCHAIN_PROFILE=notarytool npm run desktop:distnpm run desktop:audit-package
Known Caveats
- Manual installed-app updater smoke from a previous public version was not run in this session. The release still verifies packaged updater metadata and public
latest-mac.yml. - If a user installed a very old AVE build that does not contain
Contents/Resources/app-update.yml, that build may not self-update. Installing this DMG manually restores the updater path for future releases. - The retired-model cleanup targets the old Qwen3.5 2B OptiQ language model. Qwen vision 2B remains supported for visual analysis.