Highlights
- Vellum now supports conversation forking — you can branch off from any message to explore alternative directions without losing your original conversation history.
- A new LLM context inspector lets you drill into the exact prompts and responses sent to the AI model for any assistant message, including full request logs and normalized payload views.
- Conversation disk views have been introduced, projecting conversations to disk in a structured format to improve reliability, attachment handling, and cross-feature consistency.
- The inference provider system has been significantly expanded: you can now bring your own API key for OpenAI, Gemini, and other providers, with per-provider model selection and key management directly in settings.
- macOS chat scrolling and bottom-pinning have been rearchitected with a dedicated coordinator, scroll loop guard, and diagnostics to eliminate jank and runaway scroll behavior.
Features
- Add double-click zoom-to-read interaction for constellation nodes #19468
- Add native hover tooltips to constellation graph nodes #19467
- Export client diagnostics and hang artifacts #19455
- Instrument transcript scroll and anchor diagnostics #19451
- Capture hang context for prolonged main-thread stalls #19450
- Add transcript diagnostics to debug-state snapshot #19449
- Add assistant progress transition diagnostics #19446
- Add chat bottom-pin coordinator #19441
- Add chat scroll loop guard helper #19440
- Add macOS client diagnostics store and session logs #19439
- Add shared visible-message filter for chat pagination #19437
- Add pulsing glow to avatar while assistant is generating #19429
- Add line-level diff highlighting with red/green backgrounds #19428
- Add embedding config card to Models & Services settings #19394
- Add embedding config state and methods to SettingsStore #19390
- Show all LLM calls per turn in context inspector #19389
- Add helper to resolve all assistant messages in a turn #19388
- Add GET/PUT /v1/config/embeddings API endpoints #19387
- Add embedding config SettingsClient methods and Codable types #19386
- Use persisted runtime providers in LLM context #19383
- Improve LLM inspector provider labels #19382
- Add provider column to LLM request logs #19381
- Add iOS message-level fork action #19342
- Add iOS fork parent navigation and current-tip UI #19337
- Add macOS fork header action and parent link #19335
- Add macOS message-level fork action #19334
- Advertise /fork in Apple slash help #19333
- Wire iOS conversation forking into IOSConversationStore #19325
- Wire macOS conversation forking into ConversationManager #19323
- Add conversation fork route #19315
- Add storage-level conversation fork cloning #19311
- Add summary-first overview tab for LLM inspector #19308
- Render normalized response sections in LLM inspector #19307
- Render normalized prompt sections in LLM inspector #19305
- Add shared conversation fork and detail clients #19304
- Expose fork lineage in conversation reads #19303
- Convert LLM context inspector to master-detail shell #19299
- Add conversation fork transport types #19297
- Add conversation fork lineage columns #19296
- Add shared fork intent hooks #19295
- Add normalized LLM context models for Apple clients #19294
- Normalize LLM context summaries in assistant route #19293
- Add structured raw payload viewer for message inspector #19292
- Sync macOS slash picker with shared command catalog #19272
- Render command help bubble from shared slash catalog #19271
- Add shared slash command catalog for Apple chat clients #19268
- Genericize managed OAuth flow and fix disconnect error handling #19267
- Make slash command
Build: 0.5.2
Commit: 3faa497aa
Built at: 2026-03-19 13:55:58 UTC