Skip to content

Cognithor 0.92.7

Choose a tag to compare

View all tags
@github-actions github-actions released this 23 Apr 21:04
· 415 commits to main since this release
v0.92.7
7eadc60
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

What's Changed

Features

  • feat(vllm-video): native video input via Qwen3.6-27B + 5 review rounds (#140) (7eadc60)
  • feat(flutter): implement URL-einfügen dialog for video URL paste (d2ede0b)
  • feat(installer): bundle LGPL ffmpeg for Windows + CI verification step (fc4ce15)
  • feat(flutter): chat bubble renders video kind with thumbnail + long-video banner (a84f60a)
  • feat(flutter): ChatProvider.sendVideo + URL-paste video detection (b7ce496)
  • feat(flutter): chat-input paperclip becomes PopupMenuButton with Video entry (0292709)
  • feat(gateway): video attachment classification + MediaServer/Cleanup lifecycle (55a9dca)
  • feat(api): /api/media/upload + /api/media/thumb endpoints (363a14f)
  • feat(llm): UnifiedLLMClient hard-errors on video when vLLM DEGRADED (6fbf130)
  • feat(planner): route video_attachment to vision_model_detail (97db041)
  • feat(models): WorkingMemory.video_attachment per-turn field (3f95e88)
  • feat(vllm): orchestrator threads spike flags + media-url env into docker run (082f50c)
  • feat(vllm): VLLMBackend.chat(video=) with mm_processor_kwargs payload (96b3f2c)
  • feat(video): VideoCleanupWorker with session registry + mtime TTL sweep (5aaf460)
  • feat(media): MediaUploadServer FastAPI app + ephemeral-port lifecycle (5791048)
  • feat(media): MediaUploadServer save_upload with quota + LRU eviction (ee7c76d)
  • feat(config): video + launcher fields on VLLMConfig (97f6276)
  • feat(llm): MediaUploadError hierarchy for video upload failures (f5c0384)
  • feat(video): ffprobe-based resolve_sampling with adaptive + override modes (b92244b)
  • feat(video): VideoSampling dataclass + duration bucket rules (b65108e)

Fixes

  • fix(test): browser-agent not_started_error on Python 3.12+ (00496ba)
  • fix(installer): also flatten iss ffmpeg paths + ruff wrap (f6a9ef0)
  • fix(installer): flatten ffmpeg archive to fixed bin/ path (5098df4)
  • fix(flutter): promote KeyboardListener FocusNode to state field (3c218c8)
  • fix(vllm): _attach_images_to_last_user preserves list content symmetrically (4ef4c36)
  • fix(api): surface quota-exceeded recovery hint to client (338e93b)
  • fix(vllm): thread live VLLMConfig through orchestrator construction (fff7cb6)
  • fix(vllm): log docker run command + failure details for observability (e07d48d)
  • fix(flutter): dispose URL-dialog TextEditingController to prevent leak (bbf3620)
  • fix(video): VideoCleanupWorker.start() is idempotent (9c790a9)
  • fix(api): share VLLMOrchestrator between Gateway and backends_api endpoints (aad9506)
  • fix(flutter): defer _isUploading reset to post-frame callback (e42828e)
  • fix(vllm): VLLMBackend.chat_stream accepts + forwards video kwarg (558c92c)
  • fix(media): threading.Lock around save_upload to prevent quota TOCTOU (c972628)
  • fix(media): harden path-traversal guards via resolved is_relative_to check (2cc64cc)
  • fix(gateway): wire VideoCleanupWorker.on_session_close into stale-session sweep (74bfe54)
  • fix(flutter): strip query + fragment from URL-paste filename derivation (e51f3e9)
  • fix(gateway): offload _build_video_attachment to thread pool (2db7444)
  • fix(api): media upload handler offloads ffprobe + ffmpeg to thread pool (05a033b)
  • fix(flutter): block send button while video upload is in progress (b8b38c9)
  • fix(vllm): _attach_video_to_last_user preserves pre-existing list content (90dcfea)
  • fix(config): default vLLM image to cu130-nightly per spike (e082d6e)

Other

  • chore(release): bump Flutter kFrontendVersion to 0.92.7 (f38408e)
  • chore(release): sync version + flutter pubspec version to 0.92.7 (45f0cb2)
  • chore(release): bump to 0.92.7 for video-input feature (a2cecb2)
  • docs(video): document cu130-nightly rolling-tag risk + mitigation (5bba52a)
  • style: clear all repo-wide ruff errors + format (5cab95a)
  • chore(changelog): document video-input via vLLM feature (ba108cd)
  • docs(video): manual smoke-test recipes 9-15 for video input (8defc55)
  • docs(video): user guide video-input section with troubleshooting + config (b3381ae)
  • test(vllm): fake server round-trip for video_url + mm_processor_kwargs (ddfe5f1)
  • refactor(video): pin resolve_sampling mocks to module-scoped subprocess (3323f1a)
  • refactor(video): tighten VideoSampling docstring + error-message test (ca7a38e)
  • docs(spike): Day-1 video-input spike complete, plan/spec amended (cc1e21c)
  • docs(plan): video-input via vLLM implementation plan (0a6d6c6)
  • docs(spec): fix inconsistencies introduced by single-video refactor (b70bf3b)
  • docs(spec): revise video-input spec with honest risk review (0afac63)
  • docs(spec): video-input via vLLM for Qwen3.6-27B (5c5ae89)

Full Changelog: v0.92.6...v0.92.7

Assets 8
Loading