CrewHub v0.15.0: Your AI Team Just Learned to Run Meetings

AI Meetings and bot pathfinding turn your agent crew into an actual collaborative team. Structured discussions replace chaotic multi-agent chats, producing organized summaries and executable action items.

If you’re managing multiple AI agents who need to actually work together (not just talk past each other), this release is for you.

Your bots don’t just chat anymore. They physically walk to a meeting table, form a circle, and have structured round-robin discussions that produce real results.

✨ What’s New

🤝 AI Meetings: Collaboration in 3D Space

Click a meeting table in your HQ (or any project room). Pick participants and a topic. Watch what happens:

• Bots navigate through hallways and doors to reach the meeting room (no more wall-walking)
• Each agent takes organized turns building on what others said (round-robin discussion)
• Live speech bubbles show who’s speaking in real-time
• Structured output appears when the meeting ends

Bots physically walk to the meeting table and form a circle. Real collaboration in 3D space.

When the meeting completes, you get:

✅ Structured summaries organized by theme (not just raw transcript)
✅ Action items with Execute/Add to Planner buttons (from discussion to work in seconds)
✅ Documented decisions so nothing gets lost
✅ Follow-up meetings that carry context forward automatically

Action items parsed from the discussion, each with agent assignment, priority, and direct execution buttons.

Post-Meeting Workflow (6 new features):

1. Filename with context — Meeting files named by topic: 2026-02-13-sprint-planning.md
2. Sidebar panel — Fullscreen results overlay with max-width 1024px, auto-opens on completion
3. Better results UI — Structured parsing with ActionItemCard, MeetingSummaryCard, RoundAccordion components
4. Action items → Tasks — Database integration, Planner API, agent execution, SSE status tracking
5. Follow-up meetings — Parent meeting context loads automatically, pre-filled dialog
6. Meeting history browser — Paginated, tabbed interface, room-scoped search

🗺️ Spatial Awareness: Bots That Navigate Properly

Bot pathfinding implemented with waypoint-based system:

• Bots walk through room doors → hallways → target room doors → destination
• Smooth rotation at waypoints
• Fallback to straight-line if room data unavailable
• Works across multi-room layouts

No more bots walking through walls. They navigate like they understand the space.

🤖 Agent Identity Pattern

Single identity across all surfaces. Your Dev bot in the 3D world is the same Dev in chat, same Dev in Planner tasks. The 3D avatar is a representation, not a different personality.

This prevents identity drift and keeps your crew consistent across all interactions.

⏱️ “Supervising” Status

New agent status for long-running subagent tasks. When an agent spawns a subagent to work on something complex, they show as “Supervising” instead of incorrectly appearing as “Sleeping.”

Better status semantics = clearer understanding of what your crew is actually doing.

🔧 Improvements

• Meeting results UX redesign — Matches markdown viewer layout exactly: left sidebar menu, 4 views (Structured/Actions/Transcript/Raw), consistent patterns
• Dev error log toggle — F6 keybinding shows/hides error panel in dev mode, localStorage persistence
• Homepage AI Meetings section — Added to crewhub.dev after Zones System, includes GIF + screenshots + feature badges
• Demo content updated — 7 demo tasks, 2 demo meetings with transcripts, 5 demo documents covering v0.15.0 features
• Meeting tables in all project rooms — Not just HQ, every project room gets a meeting table with auto-context from the room

🐛 Bug Fixes & Technical

• Cancel meeting fix — Backend SSE broadcast + frontend state update (was double failure)
• Meeting table collision — Offset table to [2, 0.16, 2] to prevent bot overlap
• File tree visibility — Fallback + depth expansion for document selector
• Round display bug — Added num_rounds to SSE event
• Bot status “in meeting” — New status type for active meeting participants

TL;DR for Developers

• Schema v15: meetings, meeting_participants, meeting_turns, meeting_action_items, parent_meeting_id column
• Breaking changes: None (additive only)
• New statuses: 'meeting' and 'supervising' agent status types
• SSE events: 8 new meeting event types (meeting-started, meeting-round-started, meeting-turn-complete, meeting-round-complete, meeting-synthesizing, meeting-complete, meeting-cancelled, meeting-error)
• Debugging: F6 toggles dev error log, meeting state visible in SSE panel
• Architecture: MeetingOrchestrator with state machine (GATHERING → ROUND_1/2/3 → SYNTHESIZING → COMPLETE), ~3500-4000 token budget per meeting
• Pathfinding: Waypoint-based system in meetingStore.ts, room door coordinates calculated from layout
• Code: ~5,500 lines across 20+ files

Full technical docs: docs/meetings/ directory

🔜 What’s Next

Coming in v0.14.1 hotfix:
TransformControls position improvements (controls not centered on selected part)

Roadmap ahead:
Grid boundary fine-tuning, enhanced spatial awareness, markdown viewer Phase 4 polish

---

Try It Yourself

Click the meeting table in your CrewHub HQ. Watch your bots physically walk to the table, form a circle, and have an actual structured discussion.

This isn’t just a chat feature. It’s a workflow transformation.

Your AI team just learned to run better meetings than most humans.

---

Want to see the meta example? Read how we used an actual AI meeting to write the blog post about AI meetings: My AI Team Just Had a Meeting About Meetings