v1.6.0

What's Changed

• fix: resolve citations when model prepends a toolu_ prefix by @miurla in #888
• fix: handle compound TLDs in citation domain display by @miurla in #889
• fix: card pasted content by size only, not line count by @miurla in #892
• Add knowledge library notes by @miurla in #893
• Add paginated library notes loading by @miurla in #895
• Remove note save button shown event by @miurla in #896
• Retire related_q_style_v2 A/B; render follow-ups as inline link by @miurla in #897
• Use signed URLs for uploaded files by @miurla in #898
• Force Vercel runtime to Node.js by @miurla in #899
• Stabilize attachment image previews by @miurla in #900
• Make related questions conditional by @miurla in #901
• Support signed library file reattachments by @miurla in #903
• Place Add attachment menu below trigger by @miurla in #904
• perf(rls): wrap current_setting() in (select …) for per-statement evaluation by @dmitrymaranik in #902
• fix(db): correct 0016 journal timestamp and fix format CI break by @miurla in #905
• Gate related questions behind related_questions_enabled flag by @miurla in #906
• fix: handle bare string for include_domains/exclude_domains in Zod schema search tools by @simply24365 in #910
• fix(search): keep images in model-facing tool output by @miurla in #914
• Revert "Gate related questions behind related_questions_enabled flag" by @miurla in #915
• refactor: migrate Langfuse integration from langfuse-vercel to JS SDK v5 by @miurla in #917
• feat: make overall daily chat limit configurable via env by @miurla in #919
• feat(analytics): track overall chat limit checks by @miurla in #920
• fix(analytics): deliver rate-limit events via after() by @miurla in #921
• Compact historical agent messages without provider-specific pruning by @miurla in #922
• Switch cloud models to GPT-5.6 Luna by @miurla in #923
• Reduce title generation overhead by @miurla in #924
• fix(feedback): send Langfuse score once per feedback action by @miurla in #925
• fix(render): restore related-questions spec emission on the current model by @miurla in #926
• fix(streaming): log structural diagnostics for provider API errors by @miurla in #928
• fix(streaming): mark the research span as ERROR when the stream fails by @miurla in #930
• fix(streaming): keep the prompt prefix stable across turns by @miurla in #932
• fix(streaming): stop rewriting source context as turns are appended by @miurla in #933
• fix(streaming): stop recoverable tool errors from failing the research trace by @miurla in #934
• fix(streaming): mark the research span when the answer is empty by @miurla in #937
• fix(prompts): show an image block in the worked answer examples by @miurla in #938
• fix(prompts): number citations by result position, not by tool call by @miurla in #939
• fix(prompts): keep spec fences out of the worked answer example by @miurla in #940
• fix(storage): sign file URLs from a fixed window so attachments stay cached by @miurla in #941
• feat(streaming): cap how many attachments are replayed to the model by @miurla in #942
• fix(storage): treat a blank signing window as unset, not as disabled by @miurla in #943
• fix(streaming): record provider API call diagnostics on the failing trace by @miurla in #945
• fix: deduplicate repeated attachments instead of replaying every copy by @miurla in #947
• fix(errors): describe a failed tool call by its resource, not by the app by @miurla in #949
• feat(streaming): bound replayed attachments by weight, not only by count by @miurla in #951
• fix(upload): reject files whose contents are not a supported format by @miurla in #952
• fix(errors): classify a malformed provider request instead of a generic 400 by @miurla in #953
• fix(fetch): reject a tool argument that is not a URL before requesting it by @miurla in #955
• fix(tracing): distinguish unclassified stream failures from one another and from user aborts by @miurla in #956
• fix(openai): set per-chat prompt cache key by @sushanttiwari-14 in #959
• fix(streaming): keep replayed source context from splitting a surrogate pair by @miurla in #960
• fix(search): describe a failed provider request by its HTTP status by @miurla in #961
• fix(search): surface complete Brave failures by @sushanttiwari-14 in #964
• fix(tracing): record whether a stream failure happened before the first generation by @miurla in #965
• fix: stabilize related questions in deep chats by @sushanttiwari-14 in #966
• fix(prompt): keep the related-questions reminder out of replayed history by @miurla in #967

New Contributors

• @dmitrymaranik made their first contribution in #902
• @simply24365 made their first contribution in #910
• @sushanttiwari-14 made their first contribution in #959

Full Changelog: v1.5.0...v1.6.0