Repository navigation
Fix workspace replay, stale cache reads and search state; refresh image/video models - #417
Merged
Merged
Conversation
Replaying a workspace_created whose name another id already held went through INSERT OR REPLACE, which resolved UNIQUE(name) by deleting the other row. Streams from 5.17-5.20.x (and two devices creating the same name before syncing) therefore lost a workspace, its edits, and the parent row of its sessions, states, projects and tasks on cache rebuild. Both workspaces now keep their rows and ids, so every reference still resolves. The earliest-created (ties: smaller id) keeps the name; the other is shown as "<name> (<id>)", independent of shard order. A replayed rename onto a taken name applies the rest of the event with the same disambiguated name instead of failing it, and the live update refuses such a rename before appending its event. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…nced "Nexus: Rebuild cache" wiped and replayed SQLite but left the shared repository QueryCache alone, so getWorkspace (and every other repository read) answered with pre-rebuild rows until the 60 s TTL or a reload. rebuildCache now clears it once the replay settles, on failure too. Startup clears it after its sync/rebuild, since the adapter serves reads while replay is still filling the cache. And a sync() that arrives while another is in flight now waits for the follow-up run instead of returning a fabricated 0-event success, so "Refresh synced data" clears the query cache after its changes have landed, not before. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
search content decided between "off" and "initializing" from the saved setting alone. A manager turned off at runtime (setEnabled(false), or a toggle not yet saved) and a manager whose engine failed to start or whose model failed to load all answered semantic=true with "still initializing", which agents retry indefinitely. resolveSemanticSearchState now also reads the manager: turned off reads as off (keyword fallback, "turned off" note and description), and a failed start or a model-load error reads as a new 'error' state that falls back to keyword search with a note naming the cause and pointing at "Try again". "Initializing" is left for a manager that is on and starting. EmbeddingManager keeps the last start error in getStatus(). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With no image provider set up, the Images fold summary named the default
model from the catalog ("Nano Banana 2") while its Model dropdown said
"No models available". The Video fold did the same for a saved provider
without a key ("Veo 3.1 · 16:9 · 720p" next to "(unavailable)").
The summaries now read "No image provider" / "No video provider" when no
provider of that kind is set up, and "Not set" when the selected one is
unusable (no key, turned off, or listed no models). The saved model is
kept, so it applies again once its provider is set up.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Image: - Add Nano Banana 2.1 (gemini-nano-banana-2.1) on Google direct and OpenRouter, GPT Image 2.5 Sunburst/Flare via OpenRouter, FLUX.3 Image, Seedream 5.0 Pro and Seedream 5.0 Flash via OpenRouter. - Mark gpt-image-1 (shuts down 2026-10-23) and gpt-image-1.5 / gpt-image-1-mini (2026-12-01) as deprecated; gemini-2.5-flash-image's shutdown moved to 2027-03-15 per Google's deprecations page. Video: - Veo routes accept 16:9 and 9:16 only: drop 1:1 from Veo 3.1 (Google direct) and Veo 3.1 Fast (OpenRouter); add 4k to Veo 3.1 on OpenRouter. - Send OpenRouter's case-sensitive '4K' instead of the internal '4k'. - Add Seedance 2.5, Kling v3.0 Pro, Grok Imagine Video 1.5 and Runway Gen-4.5 via OpenRouter. - Google retires every Veo 3.1 preview id on 2026-10-22 and its successor needs the Interactions API, so the shipped video default moves to OpenRouter's google/veo-3.1-fast. Regenerate the tool catalogs for the image model enum, and add a nexus-model-updates protocol for video models. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Follow-ups to #416: four data-safety and correctness fixes, and an image/video model catalog refresh ahead of Google's Veo 3.1 preview shutdown on 2026-10-22.
Fixes
workspace_createdreplay usedINSERT OR REPLACE; a create with the same name and a different id hitUNIQUE(name)and SQLite deleted the other workspace's row and edits (order-dependent). Both rows are now kept under their own ids so sessions, states, traces, tasks and conversations stay linked; the earliest-created keeps the name and the other is shown asName (id), independent of replay order. Colliding replayed renames keep their other changes; live renames onto a taken name are refused before the event is written.errorstate with a clear note; "initializing" only while actually starting.Image and video models
veo-3.1-generate-preview, shuts down 2026-10-22),gpt-image-1(2026-10-23),gpt-image-1.5andgpt-image-1-mini(2026-12-01),gemini-2.5-flash-image(2027-03-15).google/veo-3.1-fast(the Google-direct default is shutting down). Existing saved defaults are unchanged.4kinstead of4Kand would have failed.gemini-omni-1.1-flash(Interactions API), MiniMax Hailuo 3 (2K only).schemas/5.20.4/*regenerated asschemas:checkrequires (image model enum).Test plan
npm run build(schemas check, lint, tsc, esbuild, connector)check_model_registry.pyexits 0 for google, openai, openrouter🤖 Generated with Claude Code