✨ Feature
- Added automatic conversation renaming from existing conversation content, available in chat titles and recent conversation rename dialogs.
- Added conversation search across navigation search and the Recent page, covering titles, labels, models, projects, and message content.
- Added admin controls for context compression, including global enablement, turn and token triggers, preserved turns, background compression, LLM compression model, and custom summary prompts.
- Added a chat context policy API so the user-facing auto-compression setting only appears when enabled by admins.
- Added model availability indicators for “No upstream” and “Not enabled” states in model management.
- Added inline protocol editing for model upstream source bindings.
- Added file-context explanations in process traces, showing whether files were sent as visual context, full text, retrieval chunks, budgeted context, or skipped.
- Added support for querying only public, routable models for model selection and model ordering.
- Added expand/collapse controls for long user messages in chat history.
⚡ Optimization
- Improved large admin tables with virtual scrolling and dynamic viewport heights across models, users, logs, tools, and upstream sources.
- Replaced multi-row skeleton loading with lighter table loading states to reduce layout jitter.
- Improved bulk action reliability by chunking deletes, archive changes, project moves, role/status updates, balance updates, and file deletion.
- Refreshed the model catalog when opening the model picker or using
@mentions, reducing stale model lists. - Improved model default option synchronization when model defaults change and the user has not customized current options.
- Limited the model ordering sheet to available public routable models so disabled or unbound models do not affect frontend ordering.
- Improved model directory ordering by prioritizing available models and keeping vendor grouping more stable.
- Improved context compression to use active branch and snapshot boundaries, avoiding duplicate summaries plus full history in prompts.
- Improved snapshot reuse for rolling context compression, allowing new summaries to build on previous summaries.
- Moved context compression settings from file/RAG settings into conversation settings for clearer admin ownership.
- Increased default full-context file budgets to 64KB and 65,536 tokens.
- Increased max page size to 1000 for announcements, audit logs, billing, users, system events, and model channels.
- Debounced table toolbar search while keeping typing and clearing responsive during loading.
- Removed direct email display from the sidebar user menu.
- Improved file-context badge tooltips with clearer descriptions.
🐛 Fixed
- Fixed OpenRouter Responses not being recognized as a lightweight model probe protocol.
- Fixed unstable blank-line normalization when HTML visual content is mixed with Markdown.
- Fixed missing failure feedback when copying external links fails.
- Fixed long user message measurement and collapse-state instability.
- Fixed unclear feedback and unstable rollback behavior in large bulk management actions.
- Fixed vendor display and grouping relying too heavily on model names; vendor fields are now preferred.