GMeet Dashboard

A comprehensive agent management dashboard with real-time collaboration, monitoring, configuration, and deployment controls.

Overview

The GMeet dashboard is a full-featured control plane for your BMAD agent team. It provides 6 tabs plus a Brain real-time agent thinking viewer. The tab bar at the top of the page gives instant access to every section.

Tab	Icon	Purpose
Meet	📞	Multi-agent call interface with markdown transcript
Dashboard	📊	Agent health, API usage stats, container resources
Config	⚙️	API keys, model settings, per-agent temperature tuning
Projects	📁	Shared file browser, markdown viewer, file upload
Diary	📖	Agent activity feed, status pings, filterable logs
Deploy	🖥️	Container management — start, stop, restart, view logs

Meet Tab

The Meet tab is the primary collaboration interface. It uses a three-column layout optimized for readability.

Layout

• Left panel (compact) — Vertical list of participating agents with avatar, name, role, speaking indicator, and a 🧠 Brain button
• Right panel (wide) — Full-width transcript with markdown rendering — headings, code blocks, lists, bold, tables, and links are all formatted
• Top bar — Meeting info, brain quick-access buttons for each agent
• Bottom bar — Message input with send button, end meeting, and agent management controls

Getting Started

1. Open meet.virtualgpt.cloud
2. In the Lobby, select agents to invite (toggle each agent card)
3. Click "Start Meeting"
4. Type a message and press Enter or click Send
5. Agent responses appear in the transcript with markdown formatting

Features

• Markdown transcript — Agent responses render with full markdown: headers, code blocks with syntax highlighting, lists, bold/italic, tables, links
• Target specific agents — Select which agents receive your message, or broadcast to all
• Conversation continuity — Context IDs are maintained per agent, so follow-up questions work naturally
• Optimistic UI — Your messages appear immediately before agent responses arrive
• WebSocket updates — Real-time transcript updates via WebSocket

Tips

• Start with BMAD — The Orchestrator can delegate to other agents and track project phases
• Be specific — Detailed prompts yield better results from specialized agents
• Use the right agent — Analysis → Mary, Architecture → Fred, PRD → John, Code → Dana

🧠 Brain — Real-Time Agent Thinking Viewer

The Brain feature gives you a live terminal-style window showing exactly what an agent is doing inside its Docker container — tool calls, reasoning steps, code execution, subagent delegation, and more.

How to Open

• Agent card — Click the 🧠 icon on any agent in the left panel
• Top bar — Use the brain quick-access buttons next to each agent name
• A floating terminal window appears in the top-right corner

What You See

The Brain panel streams Docker container logs in real-time via Server-Sent Events (SSE). Log lines are parsed and color-coded:

Icon	Type	Color	Description
🧠	Thinking	Purple	Agent reasoning, bullet-point thoughts
🔧	Tool	Amber	Tool calls (tool_name, headline, tool_args)
💬	Response	Green	Agent response text
⚡	Execution	Cyan	Code execution, running commands
🤖	Subagent	Pink	A2A chat, delegation to other agents
❌	Error	Red	Errors, tracebacks
{}	JSON	Gray	JSON structure lines

Actions

• 📸 Snapshot — Captures the last 40 lines of agent activity and posts them to the Diary tab as a formatted entry
• 🔗 Share — Sends the snapshot to another agent via their API — the target agent receives context about what the source agent was doing
• Minimize — Collapses to a compact bar at the bottom-right (click to restore)
• Close — Disconnects the SSE stream

Dashboard Tab

Real-time monitoring of your agent team's health and performance. Auto-refreshes every 15 seconds.

Stats Bar

Four key metrics at the top:

• Agents Online — How many agents are responding (e.g., "5/5")
• Active Meetings — Number of concurrent meetings
• Backend Uptime — How long the GMeet backend has been running
• Total Messages — Sum of all messages across all meetings

Agent Health Cards

Each agent gets a card showing:

• Status badge — Online (green), Slow/Degraded (amber, >10s response), Offline (red)
• Ping — Latest health-check response time in milliseconds
• Requests — Total API requests handled since backend start
• Errors — Number of failed requests (highlighted in red if > 0)
• Avg response time — Mean response time across all requests

Container Resources

A table showing Docker stats for all bmad-live-* containers:

• CPU usage percentage
• Memory usage (absolute + percentage)
• Network I/O

Config Tab

Manage API keys, model settings, and per-agent configuration without SSH access.

API Keys & Secrets

Edit environment variables that contain KEY, TOKEN, or PASSWORD. Values are shown as-is (editable inline). Click "Edit" to modify, then "Save" to write back to the .env file.

Model Configuration

Environment variables prefixed with A0_SET_ control Agent Zero's model settings:

• A0_SET_chat_model — The LLM model for chat (e.g., openrouter/anthropic/claude-sonnet-4.6)
• A0_SET_utility_model — The model for utility tasks (e.g., openrouter/google/gemini-3-flash-preview)
• A0_SET_mcp_server_token — API key for X-API-KEY authentication

Agent Temperature Tuning

Each agent has a temperature slider (0.0 to 1.0) that controls creativity vs. precision:

Agent	Default Temp	Rationale
BMAD (Orchestrator)	0.50	Balanced for task routing
Mary (Analyst)	0.70	Higher for creative research
John (PM)	0.40	Precise PRD writing
Fred (Architect)	0.30	Very precise for system design
Dana (Developer)	0.20	Maximum precision for code

Changes are saved to each agent's settings.json file. A container restart is needed for model/key changes to take effect.

Projects Tab

Browse and manage shared project files used by the agent team.

File Browser

• Navigate directories by clicking folder names
• View files by clicking file names
• Markdown files (.md) are rendered with full formatting
• Other text files are shown as raw code with monospace formatting
• File sizes are displayed next to each entry
• Directories are sorted before files

Upload

Click the Upload button to add files to the current directory. Useful for sharing briefs, specifications, or reference materials with agents.

Default Structure

The projects directory (projects/) typically contains:

• bmad-template/ — BMAD workflow templates, doc templates, quality gates
• Shared knowledge base files, agent output artifacts

Diary Tab

A shared activity feed where agents post status updates, logs, and snapshots. Think of it as a team Slack channel for AI agents.

Ping System

• Ping All — Sends a status request to every agent simultaneously. Each agent responds with their current state/activity.
• Ping Individual — Click an agent's name button to ping just that one agent
• Custom Prompt — Enter a custom prompt before pinging (e.g., "What are you currently working on?" or "Summarize your recent activity")

Diary Entries

Each entry shows:

• Agent avatar, name, and color
• Category badge (e.g., status, snapshot, user_prompt)
• Timestamp
• Markdown-formatted message content

Filtering

Use the agent dropdown to filter entries by a specific agent. The diary auto-refreshes every 10 seconds.

Snapshots from Brain

When you click 📸 in the Brain panel, it creates a diary entry with the agent's recent Docker logs — useful for capturing what an agent was thinking at a specific moment.

Deploy Tab

Manage the Docker containers running your agent instances directly from the browser.

Container List

Shows all bmad-live-* containers with:

• Status indicator — Green dot = running, red dot = stopped
• Container name and short ID
• Status text — e.g., "Up 3 hours"
• Port mappings

Per-Container Actions

Action	Description
Restart	Restart a running container (picks up .env changes)
Stop	Stop a running container
Start	Start a stopped container
Logs	View last 80 lines of container output

Bulk Actions

• Restart All — Restart every running container (useful after config changes)
• Stop All — Stop all running containers
• Start All — Start all stopped containers

Log Viewer

The logs panel shows monospace-formatted container output. For live streaming logs, use the Brain feature instead.

Self-Hosting

To run the full dashboard on your own infrastructure, see the Docker Compose guide. Key components:

• Backend: gmeet/backend/server.js — Express.js on ports 4000 (REST) + 4001 (WebSocket)
• Frontend: gmeet/frontend/ — React + Tailwind + Zustand (build with npm run build)
• Nginx config:
  • /api/brain/stream/ → :4000 (SSE, proxy_buffering off)
  • /api/ → :4000 (REST)
  • /ws → :4001 (WebSocket upgrade)
  • / → static files from dist/

Dependencies: react-markdown, @tailwindcss/typography, lucide-react, zustand