0.8.32

Preserve the AI & Search settings scroll position when changing settings.

0.8.31

Repair semantic-index partition validation for unconstrained dimensions, preserve provider-free routing across cache revisions, and automatically recover a loaded index when sidecar persistence or encoder settings temporarily prevent routing warmup.

0.8.30

Raise Open WebUI chat and prompt-response output ceilings to 4,096 tokens, explicitly carry bounded output budgets to llama-server and Ollama transports, and add live-validation/project rule coverage.

Semantic Todoist Sync 0.8.29

Automatically self-heal semantic-index queues and align Open WebUI model limits.\n\n- Schedules and retries pending semantic-index flushes so compatibility rebuilds and transient failures cannot strand vault evidence retrieval.\n- Coalesces large pending bursts into one rebuild instead of hundreds of serial per-file embedding jobs.\n- Reads nested Open WebUI/llama-server context metadata and uses the discovered model window for chat and task preflight budgets.\n- Extends the bounded generic Open WebUI execution deadline for evidence-rich local prompts.\n- Adds regression coverage for queue scheduling, nested model context discovery, and adaptive input budgets.\n\nValidated with
ode --check main.js,
ode --check scripts/openwebui-carrier-recovery-local-test.js, 114 local assertions, git diff --check, and an Obsidian smoke reload.

Semantic Todoist Sync 0.8.28

Improve hybrid semantic chat retrieval for Open WebUI-backed models.

• Combines the live query embedding with bounded lexical keyword seeds from the local semantic index.
• Reserves up to four distinctive keyword/person matches while preserving semantic ranking.
• Carries note temporal/source metadata and relevant Todoist task-reference metadata through evidence admission.
• Adds regression coverage for named-person, general-keyword, deduplication, and reservation behavior.

Validated with node --check main.js and 109 local Open WebUI regression assertions.

Semantic Todoist Sync 0.8.27

Fix llama-server-backed OpenWebUI chat failures. OpenWebUI may return Responses API SSE events (response.output_text.delta and response.completed) through its OpenAI-compatible route; the plugin now parses that completed envelope while retaining existing Chat Completions SSE and Ollama NDJSON handling. Validated against the live connected model in Obsidian and with 104 local regression assertions.

Semantic Todoist Sync 0.8.26

Fix Open WebUI model-not-found errors by preserving the callable API model id from /api/models instead of the display name. Refresh also migrates previously saved display-name selections to matching /models/ identifiers. Validated in Obsidian against the connected Open WebUI catalog and with 102 local regression assertions.

Semantic Todoist Sync 0.8.25

Preserve the selected OpenWebUI primary model when refreshing live model discovery. Unknown capability metadata no longer causes a live selected model to be replaced; explicit generation-ineligible models are still excluded. Includes a regression test and Obsidian validation.

Semantic Todoist Sync 0.8.24

0.8.24

Open WebUI model discovery fix

• Treat live Open WebUI model endpoints as authoritative instead of retaining removed or renamed model IDs from persisted catalogs.
• Repair stale generation references to an available live model after refresh.
• Keep explicitly configured manual model IDs as the intentional exception.
• Improve model-not-found guidance so users are directed to refresh discovery and choose an installed model.

Validation

• Validated the updated 0.8.24 build in Obsidian 1.13.7 against the live Open WebUI endpoint.
• Model refresh completed successfully with 15 Open WebUI models loaded.
• Targeted regression suite passed 99 assertions.

Semantic Todoist Sync 0.8.23

0.8.23

Obsidian installation fix

• Published main.js, manifest.json, and styles.css as separate release assets, as required by the Obsidian installer.
• Synchronized manifest.json and versions.json to 0.8.23.

Plugin fixes retained

• Primary and fallback model pickers remain provider-scoped, searchable by substring (minimum three characters), and scrollable across the full provider catalog.

This release supersedes 0.8.22, whose ZIP-only packaging did not satisfy Obsidian's release asset contract.