Skip to main content

v0.2.7

This page retains the complete upstream release notes, including features, improvements, performance changes and fixes. Original record. Historical controls and restrictions describe that release; follow the v0.3.3 tutorials for current operation.

Complete release notes​

Features​

  • Generalized AI sidecar architecture: Rebuilt the AI sidecar around a protocol-first, multi-provider design — a runtime control plane, credential vault, cancellation registry, and capability descriptors; a provider config loader/registry with built-in and user configs plus Alcedo image-analysis task schemas; OpenRouter and Volcengine HTTP drivers over reqwest + rustls; a C++ host service and client with an image encoder and in-flight concurrency limit; a prefill queue; DuckDB-backed AI understanding/rating storage with FTS search; a protocol-first preset model and a generic OpenAI-compatible driver; an AiCredentialStore; a standalone QML ImageAnalysisController; and persistence/EXIF-star wiring for analysis results. (7e9483f5, 5d2c8dbc, 65b87fcd, da59a47e, a5a54357, 7f52fb61, db242253, 6dc60306, 02a1f102, 5a0ecc3c, 47e28839, ee399503, cf0d6e10, 2770c83c, 280d9494, e6180db7, c2aa3fbe, b951f2f1, 4e9b95c0, 2f3f440e)
  • AI image rating: Added AI star rating with a configurable rating-severity option (Lite/Normal/XHigh) and a segmented severity slider, threaded through the proto, server, every cloud provider, the controller, and the inspector UI. (431c41c9, 47e96ab2, 78370f66, 6d4358d3)
  • Batch image analysis with schema repair: Added batch image analysis in the OpenAI-compatible provider with automatic schema repair on validation failures, JSON sanitization/extraction utilities that tolerate markdown-fenced and prose-embedded JSON, and prompt-profile-driven system prompts. (1d68ccbb, 92c9cf5b, dd42af20, 7de2cc2d)
  • Codex OAuth provider: Added an OpenAI Codex OAuth provider with streaming SSE responses, DPAPI-backed large-credential storage, and Windows proxy passthrough into the sidecar. (befd5661)
  • Background task management: Added a background task registry with a task bar and popover, an interaction-policy controller and project DB write barrier for lock-driven task admission, run-in-background for semantic generation, analysis-rendition rendering from pipeline snapshots without disturbing the live guard, and an interactive, cancellable sidecar boot for UI responsiveness. (1ed12de4, 2a6b9b11, 7bbe5ace, 672b5634, 311cd5c2, 1b8397c7)
  • Field-specific and natural-language search: Added per-field search settings and natural-language search dispatching in the global search dialog. (435ef26d)
  • Color-temperature accuracy: Generated a Planckian-locus uv table from the CIE 1931 CMF for more accurate correlated-color-temperature mapping, and populated as-shot neutral from LibRaw camera multipliers for a correct white-balance starting point. (992f628e, 79ae6d9c)
  • DuckDB FTS and macOS credentials: Added DuckDB FTS extension loading and FTS-backed AI understanding persistence with FTS/VSS extension packaging for macOS and Windows, and added macOS keychain credential handling for the AI credential store. (ffe1d9cc, bdfbb104)

UI​

  • AI provider settings UI: Added a ccswitch-style provider selector (AiProviderProfile), a dedicated AI provider settings panel, a searchable model selector, and config-panel refinements. (da44a3d7, a566ae91, d771c266, 4b887d43, 6dc35e0c, a46a107d, c70f2598)
  • Advanced analysis and inspector UI: Added the advanced content analysis launcher dialog, an image detail panel showing AI descriptions, an image/album inspector switch, context-menu actions, and new AI-feature icons. (4f5394e1, 7e06bf14, 5fe50abd, dc01ce87)
  • Background task UI: Added an IconButton component, task kind badges, a Details button for running image-analysis tasks, and run-in-background controls in the semantic generation dialog. (672b5634, 7bbe5ace)

Bug Fixes​

  • Editor and preview stability: Fixed ROI render failure after crop/rotation. (9a1d682f)
  • Cloud AI security and correctness: Hardened cloud AI calling (explicit opt-in for non-loopback binds, provider URL validation, bounded response reads, image payload limits), switched the OpenAI-compatible driver to structured tool output for image analysis, removed the max-token cap that starved reasoning models (e.g. kimi-k2.6) of output budget, and fixed schema-checking logic. (db66809c, 249474ef, ed44c13e, a3a331c4)
  • Sidecar boot: Fixed sidecar boot parameter inconsistency. (7ef39d35)
  • DuckDB extensions: Fixed DuckDB extension packing logic and Windows extension handling. (5c91d542, 91008037)
  • macOS packaging: Fixed macOS signing and removed unsupported macOS models. (5a1424ec, c840749a)
  • Background task and AI dialog fixes: Fixed the incomplete phase 1/2 background-task implementation and the AI dialog ask logic. (d12c02a6, 91008037)