Transform MI from single-shot chat into an agentic system:
- LiteLLM provider with Claude Sonnet/Haiku models via proxy
- Agentic loop (max 5 turns): stream → parse actions → execute server-side → feed results back
- Server-side media generation (fal.ai + Gemini) as first-class MI actions
- Module data queries (rnotes, rtasks, rcal) read directly from Automerge
- System prompt enriched with recent notes, open tasks, and calendar events
- Client handles new NDJSON types (turn, action-start, action-result)
- Extracted shared media helpers, refactored image/video endpoints
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Progressive approval workflow: paste brief → AI extracts structure →
AI generates per-platform posts → review with per-post regen →
commit (saves campaign, creates threads, drafts newsletters, builds workflow DAG).
Includes MI integration for Cmd+K campaign creation and vite build entry.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add MiCanvasBridge for deep canvas state awareness (shapes, selection, viewport)
- Add MiActionExecutor to create/update/delete/move/connect shapes from MI responses
- Add MI action parsing (create-shape, connect, update-shape, delete-shape, move-shape, transform)
- Add selection transforms (align, distribute, arrange, match-size)
- Add tool suggestion schema for contextual MI hints
- Enhanced MI system prompt with action markers and transform commands
- Richer canvas context in /api/mi/ask (positions, connections, viewport, shape groups)
- Refactored tab-bar I/O chips for cleaner feed port rendering
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Jeff Emmett