What's Changed
- chore(version): bumping version to 2.0.12dev by @qbc2016 in #3507
- fix(webui): include full calendar days in schedule date filter by @pei711 in #3504
- fix(model): normalize nullable usage counters in adapters by @xiong-wei1 in #3453
- fix(tool): keep resolvable annotations when one hint fails by @mengchen-atx in #3484
- fix(app): remove server-path skill import endpoint by @qbc2016 in #3536
- fix(tts): keep CosyVoice cold-start buffer when the send fails by @Kiko0301 in #3535
- fix(embedding): deep-copy DashScope multimodal cache identifier by @Kiko0301 in #3533
- fix(embedding): honor embedding_cache on the Gemini multimodal path by @Kiko0301 in #3529
- fix(embedding): honor per-call api_key on the DashScope text path by @Wilson0511 in #3465
- fix(embedding): let a per-call dimensions override the OpenAI default by @Wilson0511 in #3468
- feat(formatter): make OpenAI chat message names match the API pattern by @kjgggggg in #3548
- fix(tts): use first Gemini audio candidate by @Kiko0301 in #3540
- feat(model): make DeepSeek thinking_enable tri-state so unset no longer force-sends thinking.type=disabled by @konlue in #3398
- feat(model): only send think option to Ollama when explicitly configured by @Kiko0301 in #3510
- feat(model): make DashScope thinking_enable tri-state by @konlue in #3192
- feat(model): only send thinking option to Moonshot when explicitly configured by @Kiko0301 in #3521
- feat(model): omit Gemini thinking_config unless thinking is explicitly configured by @Kiko0301 in #3523
- feat(model): support tri-state thinking controls across adapters and UI by @qbc2016 in #3557
- fix(message): return an empty string when no text blocks are present by @qbc2016 in #3565
- fix(schedule): respect selected end dates when creating schedules by @golitter in #3547
- fix(workspace): make the bubblewrap smoke probe require --clearenv by @RerankerGuo in #3550
- fix(embedding): preserve DashScope response input order by @lll-hhh in #3564
- fix(rag): preserve source blobs when knowledge base deletion fails by @Kiko0301 in #3556
- fix(tool): validate read offset and limit at runtime by @SSK988I in #3297
- feat(tts): add voice profile management with cloud TTS voice cloning by @qbc2016 in #2162
- fix(tool): match Bash permission rules against every subcommand of a compound command by @Kiko0301 in #3513
- fix(rag): reject nonpositive Milvus batch sizes by @shuxiachai in #3343
- fix(workspace): read SKILL.md as utf-8-sig in every skill path by @Tong-bit-art in #3177
- fix(tool): keep Edit from rewriting non-UTF-8 files lossily by @fancyboi999 in #3322
- fix(workspace): validate SKILL.md front matter in the shared installers by @Tong-bit-art in #3370
- feat(webui): add editable server-side chat queues by @qbc2016 in #2236
- fix(scheduler): honor standard cron day alternatives by @ai-yang in #3602
- fix(credential): register TypeSafeCredential in CredentialFactory by @ffffff55 in #3625
- fix(utils): apply the NaN guard on the schema-less parse path too by @ffffff55 in #3626
- fix(tool): require every subcommand to be a filesystem command for ACCEPT_EDITS auto-allow by @Kiko0301 in #3519
- fix(web-ui): reset document panel when switching knowledge bases by @yang0228 in #3422
- fix(word): preserve inline tabs and breaks in Word paragraphs by @Xscaperrr in #2840
- fix(tts): surface post-cold-start streaming_call failures instead of dropping the delta by @konlue in #3621
- fix(schedule): serialize reconciliation so stale reads cannot restore jobs by @Tong-bit-art in #3608
- fix(schedule): respect selected start times when creating schedules by @golitter in #3610
- fix(rag): preserve paragraph boundaries inside Word text boxes by @12kkqq in #3415
- fix(embedding): reject negative file cache budgets by @its-amann in #3582
- fix(middleware): retire the budget hint at a new reply boundary by @shi-felix in #3599
- fix(embedding): honor configured DashScope text parameters by @HootRock in #3367
- fix(tool): report Write failures as error chunks by @Dengzi555 in #3627
- fix(rag): preserve literal metadata values in MongoDB listings by @Jackkp0t in #3591
- fix(pipeline): clear the leader's parked TeamAssign when a member's model fails by @Java123456com in #3581
- feat(moonshot): support kimi k3.1 pro variants by @YingchaoX in #3578
- fix(app): guard session and SOP deletion side effects by @Kiko0301 in #3563
- feat(agent): add steering support to the SDK and WebUI by @qbc2016 in #3638
- test(message): cover serialization and content selection contracts by @yaxiaodang in #3494
- fix(realtime): retire unbound local user turns when the transport dies by @Wilson0511 in #3491
- fix(formatter): keep text before media in Anthropic multi-agent history by @Kiko0301 in #3525
- fix(middleware): preserve top_p=0 in tracing request attributes by @Kiko0301 in #3520
- fix(channel): degrade Discord chat kind/name lookups to contract defaults by @mosifu in #3483
- fix(model): count cached prompt tokens in anthropic input_tokens by @zhaoyta in #3477
- test(tool): add regression coverage for bash read-only permissions by @indexGin in #3474
- fix(model): take the cumulative input and cache usage of the anthropic message_delta by @pcbeingused333 in #3428
- fix(ollama): implement _get_disable_thinking_kwargs so structured-output fallback can disable thinking by @konlue in #3588
- fix(agent): count rounds completed during HITL resumption toward max_iters by @golitter in #3645
- fix(app): clean up wakeup startup failures after readiness by @lllluull in #3634
- feat(app): make the tool offload timeout configurable via create_app by @shi-felix in #3606
- fix(schema): decode escaped definition reference names by @cxyyy66 in #3000
- fix(utils): remove dangling discriminators when flattening schemas by @lll-hhh in #3618
- feat(formatter): support provider-specific media parameters in DataBlock by @qbc2016 in #2103
- refactor(model): mark the retiring Claude Sonnet 4.5 model card deprecated by @pcbeingused333 in #3429
- fix(memory): extend the frontmatter read to the closing fence when the byte cap cuts it off by @konlue in #3617
- fix(agent): fit truncated tool text to the counted token limit by @Starchaser1024 in #3441
- feat(asr): add browser speech input and DashScope models by @qbc2016 in #3298
- fix(pipeline): scope the goal pipeline retry budget to one output generation by @Wilson0511 in #3463
- fix(tool): group streamed data blocks by id and type in append_chunk by @Wilson0511 in #3457
- fix(message-bus): trim the redis replay log exactly when a cursor is given by @ZhenHaoFu810 in #3409
- fix(channel): propagate reply stream subscription failures by @wuwenbo0626 in #3374
- fix(app): clean up source blob when document registration is cancelled by @rome-xi in #3657
- fix(model): derive anthropic thinking budget from per-call max_tokens by @BlueX888 in #3292
- fix(formatter): preserve HintBlock in multi-agent requests by @zhengguangzhuo in #3321
- fix(tool): skip a PowerShell candidate that cannot run by @Kiko0301 in #3699
- feat(formatter): add audio input support for Volcengine by @qbc2016 in #3706
- fix(tool): prevent CPython startup hangs in exec_shell on Windows by @SSShooter in #3642
- fix(memory): start the ReMe app once for concurrent callers by @Kiko0301 in #3687
- fix(agent): copy the context config before applying the deprecated buffer override by @Kiko0301 in #3681
- fix(app): reject unusable saved context ratios by @ZxlDragonDoctor in #3357
- fix(app): clean up background tasks after registration faults by @Cordis798 in #3387
- fix(realtime): release streams when local transport start fails by @Kiko0301 in #3659
- fix(app): ignore duplicate index tasks for ready documents by @Smilewithoutfalling in #3658
- fix(tool): preserve empty lines in grep results by @qkycir in #3842
- fix(workspace): drain cancelled skill copies before releasing lock by @qkycir in #3844
- fix(rag): drop a leading byte order mark in TextParser by @shuxiachai in #3745
- fix(formatter): preserve image-only messages and history prompts in Ollama multi-agent formatter by @zuizhuai in #3790
- fix(mcp): refresh cached tools on tools/list_changed notification by @DingNova in #3734
- fix(middleware): default mem0 search_memory limit to top_k by @Kiko0301 in #3720
- feat(service): expose background tasks via REST endpoints by @XiuShenAl in #1865
- fix(app): fail wakeup dispatcher startup when subscription dies early by @MannXo in #3632
- fix(rag): load milvus collection before reads by @WuDYouOu in #3433
- test(app): fix race in background task renewal fairness test by @qbc2016 in #3890
- feat(rag): accept the standard YAML media type by @12kkqq in #3416
- fix(mcp): reject overlapping stateful connection attempts by @97Three in #3795
- fix(storage): write redis record value and ttl in one command by @MannXo in #3806
- fix(message-bus): refresh in-memory queue ttl on push by @haotianqi666 in #3420
- fix(hub): rewrite and declare MCP registry URL template variables by @indexGin in #3810
- fix(tool): reject invalid glob helper output by @twei43846-afk in #3038
- fix(rag): support null Qdrant metadata filters by @neu-hsc in #3055
- fix(rag): validate that the hint template is renderable by @aniruddhaadak80 in #2965
- fix(rag): accept a 12-byte webp header in the media type sniffer by @aniruddhaadak80 in #2934
- fix(embedding): make the file cache key independent of dict order by @aniruddhaadak80 in #2936
- fix(middleware): start a no-input reply on the agent's own model by @sxh313 in #2847
- fix(app): answer 422 when a session patch cannot be represented by @lihongyuan99 in #2745
- fix(app): validate channel updates instead of copying nulls into storage by @lihongyuan99 in #2744
- fix(workspace): honor an explicit max_live_stateful_mcps of 0 by @Lesereingrape in #2741
- fix(embedding): isolate OpenAI caches by API endpoint by @fhgffy in #3142
- fix(embedding): join sibling batches after failure by @97Three in #3148
- fix(app): allow WebRTC sessions to close during startup by @kassdinzheng123 in #3190
- feat(webui): display background tool status and execution duration by @qbc2016 in #3883
- fix(formatter): keep media carried by assistant messages by @lianbie in #3892
- fix(formatter): preserve anthropic tool error status by @wang1299 in #3905
- fix(middleware): release queued output after tool offload by @qkycir in #3887
- fix(model): decode the effective openai audio format by @wang1299 in #3907
- fix(ollama): close the delegated provider stream by @wang1299 in #3906
- fix(rag): reject a non-positive top_k in search by @lianbie in #3876
- feat(realtime): add GPT Live Voice Integration with Chat Agent Delegation by @qbc2016 in #3272
- fix(mcp): fence discovery responses after tool list invalidation by @qkycir in #3889
- fix(model): preserve schema definitions in model cards by @30bef100w in #3774
- fix(tool): recognise paths below a configured filesystem root by @Kiko0301 in #3719
- fix(formatter): keep parallel Ollama tool calls on one assistant message by @MaxFreedomPollard in #3910
- fix: add lower bounds to middleware and MCP config fields by @zuizhuai in #3783
- fix(embedding): reject nonpositive batch sizes by @PointerL in #3451
- fix(rag): reject negative column indices when naming excel columns by @indexGin in #3874
- fix(memory): remove obsolete ReMe dream step by @its-amann in #3851
- feat(app): mark offloaded tool placeholders with structured metadata by @konlue in #3853
- fix(rag): measure legacy S3 blobs in their URI bucket by @haotianqi666 in #3598
- fix(hub): reject timestamp skill versions carrying an invalid time by @indexGin in #3873
- fix(workspace): resolve local search paths from the workspace directory by @tomsen02 in #2637
- feat(tts): support Gemini 3.8 Flash and Flash-Lite TTS by @qbc2016 in #3677
- fix(_utils): keep nested arguments that schema-guided repair drops by @Linxiushen in #2712
- feat(model): add opt-in content-block error normalization by @qbc2016 in #3324
- fix(tool): reject Bash/PowerShell timeout below 1ms by @Li-john1021 in #3309
- fix(tool): ignore user ripgrep config in Grep by @ydflow in #3358
- fix(tests): import patch in grep tool tests by @qbc2016 in #3954
- fix(formatter): send failed gemini tool results under the error key by @David-Wu1119 in #3947
- feat(realtime): support Gemini 3.8 Live Extended Thinking by @qbc2016 in #3700
- fix(rag): normalize scores using the Qdrant collection metric by @qkycir in #3958
- feat(model): add Gemini 4 Argon and Carbon support by @qbc2016 in #3675
- fix(formatter): revert xAI assistant media promotion by @qbc2016 in #3965
- fix(app): stop cancelled background renewal workers by @qbc2016 in #3968
- feat(app): add client-declared external tools and interactive user input by @qbc2016 in #2394
- fix(realtime): restart the downlink pump after it crashes by @konlue in #3732
- fix(webui): re-measure the chat composer when its width changes by @2SinceJuly in #3953
- fix(model): surface failed non-streaming responses by @zhangkom in #3743
- fix(workspace): close MCP clients after failed gateway discovery by @qkycir in #3838
- fix(asr): join recognised sentences without inventing spaces by @Kiko0301 in #3762
- fix(app): keep failed steering items out of the queue pump by @Tong-bit-art in #3747
- fix(tool): infer default names for callable objects by @ydflow in #3963
- docs(readme): sync the NEWS section with docs/NEWS.md by @DZH166 in #3978
- fix(workspace): accept workdirs under a filesystem-root basedir by @zuizhuai in #3966
- feat(tts): add MiniMax speech synthesis by @octo-patch in #2298
- feat(model): update model catalog and sort by release date by @qbc2016 in #3997
- fix(middleware): keep other knowledge bases' context when one fails by @2SinceJuly in #3976
- fix(steering): preserve inputs across uncertain acknowledgements by @qbc2016 in #4061
- fix(embedding): join sibling multimodal batches after failure by @coderbruis in #4039
- fix(middleware): scope budget hints to the current reply by @Kiko0301 in #4043
- test(app): respect background task shutdown budget by @qbc2016 in #4063
- chore(version): bumping version to 2.0.12 by @qbc2016 in #4059
New Contributors
- @pei711 made their first contribution in #3504
- @xiong-wei1 made their first contribution in #3453
- @mengchen-atx made their first contribution in #3484
- @Kiko0301 made their first contribution in #3535
- @Wilson0511 made their first contribution in #3465
- @kjgggggg made their first contribution in #3548
- @konlue made their first contribution in #3398
- @golitter made their first contribution in #3547
- @lll-hhh made their first contribution in #3564
- @SSK988I made their first contribution in #3297
- @ai-yang made their first contribution in #3602
- @Xscaperrr made their first contribution in #2840
- @12kkqq made their first contribution in #3415
- @its-amann made their first contribution in #3582
- @Dengzi555 made their first contribution in #3627
- @Jackkp0t made their first contribution in #3591
- @Java123456com made their first contribution in #3581
- @mosifu made their first contribution in #3483
- @zhaoyta made their first contribution in #3477
- @indexGin made their first contribution in #3474
- @pcbeingused333 made their first contribution in #3428
- @lllluull made their first contribution in #3634
- @ZhenHaoFu810 made their first contribution in #3409
- @rome-xi made their first contribution in #3657
- @SSShooter made their first contribution in #3642
- @Cordis798 made their first contribution in #3387
- @Smilewithoutfalling made their first contribution in #3658
- @qkycir made their first contribution in #3842
- @zuizhuai made their first contribution in #3790
- @DingNova made their first contribution in #3734
- @WuDYouOu made their first contribution in #3433
- @97Three made their first contribution in #3795
- @haotianqi666 made their first contribution in #3420
- @kassdinzheng123 made their first contribution in #3190
- @lianbie made their first contribution in #3892
- @wang1299 made their first contribution in #3905
- @30bef100w made their first contribution in #3774
- @PointerL made their first contribution in #3451
- @tomsen02 made their first contribution in #2637
- @Linxiushen made their first contribution in #2712
- @Li-john1021 made their first contribution in #3309
- @David-Wu1119 made their first contribution in #3947
- @2SinceJuly made their first contribution in #3953
- @zhangkom made their first contribution in #3743
- @DZH166 made their first contribution in #3978
- @coderbruis made their first contribution in #4039
Full Changelog: v2.0.11...v2.0.12