Skip to content

Latest commit

 

History

History
301 lines (221 loc) · 11.8 KB

File metadata and controls

301 lines (221 loc) · 11.8 KB

User guide — AITEAM-X

Version: 2.2
Updated: 2026-04-03
Audience: AITEAM-X dashboard users


1. Introduction

AITEAM-X is a visual panel to monitor and interact with AI agents in real time. Agents appear as pixel-art characters in a virtual office with shared rooms and individual workspaces.

What you can do

  • See status and location of all 14 team agents
  • Chat directly with any agent in a real-time window
  • Enable Party Mode for sessions with all agents at once
  • Filter agents by status (Working) or module (BMGD / BMM / CORE)
  • Track sprint progress on the bottom timeline
  • Manage persistent agent memories in the Memory Vault
  • Choose the LLM model used for all interactions
  • Open the Observability page (Help menu) to inspect health, recent platform events, conversation summaries, and optional browser console capture (localhost dev)

2. Interface

2.1 Layout overview

┌──────────────────────────────────────────────────────────────┐
│ AITEAM-X                     ♫  ♪  🧠         TIME 04:13 PM │ ← Top bar
├──────────────────────────────────────────────────────────────┤
│ MODEL ▼  │  STATUS All (14/14) ▼  │  MODULE All ▼           │ ← Model bar + filters
├────────────────────────────────────────┬─────────────────────┤
│                                        │                      │
│  ┌─────────────────────────────────┐   │  ┌────────────────┐ │
│  │  JARVIS OFFICE (6)              │   │  │  KITCHEN (2)   │ │
│  │  ⚡ 6 agents collaborating      │   │  │  ☕ Mary & Sally│ │
│  │  [BMad][Builder][Link][Winston] │   │  │  chatting      │ │
│  │  [Amelia][Murat]                │   │  └────────────────┘ │
│  └─────────────────────────────────┘   │                      │
│                                        │  Workspaces grid:    │
│  ┌────┐ ┌────┐ ┌────┐ ┌────┐ ┌────┐  │  ┌───┐┌───┐┌───┐    │
│  │ CD │ │ SS │ │ MX │ │ JN │ │ BB │  │  │ WS││ WS││ WS│    │
│  └────┘ └────┘ └────┘ └────┘ └────┘  │  └───┘└───┘└───┘    │
│                                        │                      │
├─────────────────────────────────┬──────┴─────────────────────┤
│ 🎭 🧙 🔧 🏛 🎲 🕹 🎯 📊 ...   │ SPRINT ████████████ 100%  │ ← Bottom bar
└─────────────────────────────────┴─────────────────────────────┘

2.2 Top bar

  • Title “AITEAM-X” on the left — product identity
  • Music controls (♫ ♪) — toggle ambient music
  • Memory Vault (🧠) — open memory management
  • Help (?) — setup, docs, bug report, Observability (📡) — health snapshot, event log, filters, link to memory dashboard
  • Clock — local time on the right

2.3 Model bar and filters

Below the top bar there are three controls:

Control Role Options
MODEL LLM for all interactions Default (Cursor), Claude Sonnet, etc.
STATUS Filter by activity — live counts All (N/N) / Working
MODULE Filter by area All / BMGD / BMM / CORE

Filters only change on-screen visibility. Filtered agents keep running normally.

2.4 Main content

The center shows rooms and workspaces:

Shared rooms:

  • Jarvis Office — main collaborative space. Shows how many agents are present and activity text (e.g. “6 agents in collaborative session”)
  • Kitchen — break area. Agents may show context text (e.g. “Mary and Sally are chatting”)

Individual workspaces (grid):

  • Each cell is one agent’s desk
  • Shows agent name and module (BMGD / BMM / CORE)
  • Color cue: red = BMGD, blue = BMM, green = CORE
  • Click the cell to open chat

2.5 Bottom bar

  • Compact avatars for all agents — click to open chat
  • Sprint timeline — markers per completed story, with percentage
  • Sprint counter — N/N format (e.g. 14/14) and percent

3. Interacting with agents

3.1 Open chat (single click)

  1. Click any agent (room, workspace, or bottom bar)
  2. A chat window opens with prior history
  3. Type in “Reply…”
  4. Press Enter or click ▶ to send

3.2 Multi-select (Ctrl+click)

  1. Hold Ctrl and click several agents
  2. A command popup appears
  3. Type the command sent to all selected agents at once
  4. Each agent answers in their own chat window

3.3 While the agent responds

  1. Thinking bubble (💭) on the avatar
  2. Thinking text in the window — model reasoning
  3. Streaming reply — text appears in real time
  4. Timer shows elapsed time since send

3.4 Chat window

Element Description
Header Agent name + close (✕)
History User messages (right) and agent (left) with Markdown
Text field “Reply…” input
Stop Cancel in-flight generation
Handles Resize via edges and corners

3.5 Window management

Action How
Close Click ✕ — history kept, memory saved
Reopen Click the agent again
Bring forward Click the window or agent
Resize Drag edges or corners
Select text Click and drag inside the chat

Note: Closing with ✕ triggers the memory cycle: saves a checkpoint and queues the conversation for automatic LLM memory extraction.


4. Workflows

4.1 Party Mode

Party Mode puts all 14 agents in one chat for collaborative brainstorming.

To start:

  1. Open chat with BMad Master
  2. Send: party mode (or *party-mode, or 4)

Result:

  • All agents move to the Conference Room visually
  • The agent lists all 14 participants
  • You can ask questions and several agents answer in turn

To stop:

  • Send: exit party mode (or *exit, or 5)

4.2 Other workflows

Agents support the main workflows from the pack (manifests under aiteam-x-agents/ or legacy bmad/):

Workflow Lead agent Description
Product Brief Mary (Analyst) Product brief
PRD John (PM) Detailed product requirements
Architecture Winston (Architect) Technical architecture
Sprint Planning Bob (Scrum Master) Sprint planning and estimates
Dev Story Amelia (Developer) Story implementation
Code Review Bob (Scrum Master) Code review
Document Project Paige (Tech Writer) Technical documentation
Game Design Samus Shepard (Game Designer) GDD and game design

5. Memory Vault

The Memory Vault is the central panel for agent memories. Open it with 🧠 on the top bar.

5.1 Navigation

  • Agent list (left, AGENTS): pick an agent
  • Categories (left, CATEGORIES): each shows entry count
    • decisions — technical or project decisions
    • lessons — fixes, insights, learnings
    • tasks — checklist-style action items
    • projects — project context and goals
    • handoffs — summaries of prior sessions
  • Search (top): BM25 search in the selected agent’s vault
  • Right panel: entries for the selected category with date, tags, content

5.2 Creating entries manually

  1. Select agent and category
  2. Click + New entry
  3. Enter content (Markdown and #tags supported)
  4. Save — injected in that agent’s next session

5.3 The 🤖 llm badge

LLM-extracted entries show 🤖 llm for 10 minutes:

  • Review content while the badge shows
  • Veto bad entries with — removed before injection
  • After 10 minutes the badge goes away and the entry becomes permanent

5.4 How memory is injected

When a new conversation starts, the system:

  1. Finds relevant decisions and lessons (BM25)
  2. Includes the last session summary (handoff)
  3. Lists open tasks ([ ]) for the agent
  4. Adds global project context (_project.md)

This happens automatically — you do not need to do anything.


6. Visual indicators

6.1 Agent status

Indicator Meaning
Colored workspace border Green = CORE, Blue = BMM, Red = BMGD
Room status text e.g. “Orchestrating workflows”, “Deep coding”
💭 Thinking bubble Agent processing
Room counter Agents present (e.g. “JARVIS OFFICE (6)”)

6.2 Sprint timeline

  • — sprint start
  • — completed story (each marker = one story)
  • — current sprint
  • 👾 / 💀 — sprint review marker
  • N/N — completed / planned stories
  • 100% — completion percentage

6.3 Rooms and location

Place When agents are there
Jarvis Office Default active work
Kitchen Break or light tasks
Individual workspace Focused solo work
Conference Room During active Party Mode

7. Shortcuts and tips

Action How
Open chat Click agent
Multi-select Ctrl + click
Close chat ✕ (history and memory kept)
Reopen chat Click agent again
Bring window forward Click window or agent
Stop generation Stop (■)
Resize window Drag edges/corners
Select text Click and drag in chat
Send message Enter in input
Party Mode Send “party mode” to BMad Master
Exit Party Mode Send “exit party mode” to BMad Master
Open Memory Vault 🧠 on top bar
Open Observability Help (?) → Observability 📡 — or /en/observability / /pt/observability
Change LLM MODEL on filter bar
Filter by status STATUS on filter bar
Filter by module MODULE on filter bar

8. FAQ

Why is the agent slow?
Complex replies can take 10s–2m depending on task and model. The timer shows elapsed time. Use Stop to cancel.

Can I talk to several agents at once?
Yes. Each has their own window. Ctrl+click to broadcast one command to many.

What happens when I close chat?
The memory cycle runs: final checkpoint saved and conversation queued for LLM extraction. New memories may appear in the Vault with 🤖 llm.

Do agents remember past chats?
Yes. The system injects last session handoff, relevant decisions and lessons, and open tasks at the start of each new chat.

Wrong memories in the Vault?
Open Memory Vault (🧠), find the entry, delete it. If 🤖 llm is still showing (<10 min), use to veto.

How do I change the LLM?
Use MODEL on the filter bar. Applies to all agent chats from then on.

Can I move the chat window?
Position is automatic; resize by dragging edges or corners.

What is _project.md?
Shared context for all agents. You can edit it from the Vault (special “project” agent). Put stack, conventions, sprint goals anything every agent should know.

STATUS filter shows 0/0?
Data may still be loading. Wait a few seconds — polling runs every 8s. If it persists, reload the page.