NeuCharFramework (NCF)
  • NCF - NeuCharFramework
  • Projects

    • Preparation
    • NcfPackageSources Source Guide
    • DynamicWebApiEngine
    • Deployment
    • MCP (Model Context Protocol) Module
    • Senparc.AI
  • AIKernel Local Fine-Tuning
  • Help

    • Online Resources
    • Q&A Community
    • QQ Group (147054579)
    • Senparc WeChat SDK
  • Gitee
  • GitHub
  • English
  • 简体中文
GitHub
  • NCF - NeuCharFramework
  • Projects

    • Preparation
    • NcfPackageSources Source Guide
    • DynamicWebApiEngine
    • Deployment
    • MCP (Model Context Protocol) Module
    • Senparc.AI
  • AIKernel Local Fine-Tuning
  • Help

    • Online Resources
    • Q&A Community
    • QQ Group (147054579)
    • Senparc WeChat SDK
  • Gitee
  • GitHub
  • English
  • 简体中文
GitHub
  • Source Development

    • NcfPackageSources Beginner Quickstart (60 Minutes)
    • NcfPackageSources Source Guide
    • Project Relationships, Synchronization, and Release
    • Version Upgrade Notes
    • NcfPackageSources_Include_NcfSimulatedSite.sln Project Map
  • XNCF Source Analysis

    • NCF Capability Source Deep Dive (Practical)
    • XNCF Extension Library Guide (Senparc.Xncf.xxx)
    • XNCF Module Documentation Map
    • Senparc.Web and Admin
    • AIKernel Local Fine-Tuning
    • XncfBuilder Isolated Development: Source Analysis
    • NeuChar Workflow: Operations and {{= ... }} Expressions
    • Senparc.Xncf.Sandbox Environment Setup
  • NCF Library Source Analysis

    • Senparc.Ncf.Core
    • Senparc.Ncf.Database (Database Foundation)
    • Senparc.Ncf.Repository
    • Senparc.Ncf.Service
    • Senparc.Ncf.SMS
    • Senparc.Ncf.Mvc.UI
    • Senparc.Ncf.Log
    • Senparc.Ncf.Utility
    • Senparc.Ncf.XncfBase
    • Senparc.Ncf.AreaBase
    • Senparc.Ncf.DatabasePlant
  • Core Interface Source Analysis

    • IXncfRegister Interface (Current)

AgentsManager 3D Status View

Checked against the NcfPackageSources development line on 2026-08-22. This page documents the visual encoding, data boundaries, and troubleshooting path for the AgentsManager home-page 3D view. The 3D view is an operational overview, not a replacement for Agent, Group, Task, or HIL detail pages.

1. Purpose

The 3D view places local Agents, remote A2A Agents, Agent groups, and current task relationships in one scene:

  • Pillars represent Agent groups.
  • Agent spheres represent local or remote A2A Agents.
  • Links represent group membership.
  • Flowing dots represent active collaboration links.
  • Top overview shows local Agents, remote A2A, groups, active tasks, and published A2A counts.
  • Hover details show Agent state and skills; hovering or clicking a Group focuses and locks its members.

The data comes from ChatGroupAppService.GetAgentGraphSnapshot. The snapshot contains status, counts, relationships, and display-only skill kinds. It does not return Prompt bodies, conversation content, secrets, or full tool arguments.

2. Group pillar height

Pillar height is a combined “task scale + current activity” indicator. It is not a performance, quality, or model-capability score.

The current calculation is:

totalTasks = waiting + chatting + paused + finished + cancelled + failed
heightScale = 0.72 + min(1.45, totalTasks * 0.08 + runningTaskCount * 0.16)
FieldMeaning
waitingTasks waiting to execute
chattingTasks currently chatting or executing
pausedPaused tasks, including tasks waiting for human handling
finishedFinished tasks
cancelledCancelled tasks
failedFailed tasks
runningTaskCountTasks currently in Waiting, Chatting, or Paused

A group with more historical tasks is therefore taller, while current active tasks have a larger weight. To determine whether a group is actively running, combine height with color, flow dots, and the top overview instead of reading height alone.

3. Group state styles

StateVisual encodingMeaning
DisabledMuted red-gray pillar with low opacityThe group should not receive new tasks or be used as an available Workflow object
IdleGray-blue pillarNo Waiting, Chatting, or Paused tasks
WaitingBlue pillarThe group has waiting tasks
ChattingCyan pillar and flowing linksThe group has Chatting tasks
PausedOrange pillar and orange top ringThe group has paused tasks but no pending HIL request
HIL waitingMagenta pillar, pulsing top ring, and HIL waiting countA paused task has a pending humanTurn or toolApproval request

HIL detection uses the current Host process's HumanInTheLoopRequestStore. Paused alone means only that a task is paused; the magenta HIL style appears only when a pending HIL request exists.

4. Agent skill markers

Agent labels keep only the name, type, current state, and skill abbreviations. Long Prompts, scores, and runtime statistics are intentionally removed from the 3D labels. Hovering an Agent shows the full skill names in the top detail strip.

Short formSkill3D marker
FFunctionRenderBlue cube
WWorkflowGreen ring
PLegacy-compatible PluginPurple octahedron
MMCPOrange cylinder
A2APublished or remote A2AGold polyhedron
HHuman participantPink marker

Skill sources are:

  • structured bindings or legacy comma-separated Plugin names in AgentTemplate.FunctionCallNames;
  • whether McpEndpoints is configured;
  • whether the Agent is published as A2A;
  • whether the Agent is the system Human participant;
  • remote Agents default to the A2A skill.

These markers describe the available capability catalog. They do not prove that the capability was called in the current task. Actual Function execution still requires tool-invocation logs, task history, and HIL approval evidence.

5. Agent spatial layout

When a task starts, Agents are no longer compressed into one small circle next to the pillar:

  1. Members of each Group are placed in deterministic rings around the Group.
  2. Larger member sets automatically use a second or third ring.
  3. Active Groups use a larger ring radius so spheres and labels do not sit on top of the pillar.
  4. Agent labels are offset toward the outside of the ring and limited to a compact maximum of three lines.
  5. More detailed state and skill information appears in the top strip only while an Agent is hovered.

This preserves the relationship between an Agent and its Group while reducing sphere, pillar, and label collisions. The scene remains rotatable, zoomable, and Group-lockable; for large or narrow layouts, use the Group/status filters and quick-jump controls.

6. Reading order

  1. Start with the top overview to understand local Agent, remote A2A, Group, and active-task counts.
  2. Use pillar color and the top status ring to locate paused or HIL-waiting Groups.
  3. Follow flowing dots to identify active member relationships.
  4. Hover an Agent for its state and skills; hover a Group to focus its members.
  5. Click a Group to lock it, then use filters or quick-jump controls to open the detailed page.

7. Data and deployment boundaries

  • The 3D snapshot is a polled operational view, not a complete historical report.
  • Pillars include historical task status counts, so a pillar may remain tall when no task is currently running.
  • Pending HIL requests currently live in Host process memory. After a restart, without sticky multi-instance routing, or when a request reaches another instance, 3D may show database Paused without showing HIL waiting.
  • The 3D view does not prove that a model call succeeded or that a Function, Workflow, or A2A capability actually executed.
  • A2A connection state should be confirmed through the Remote Agent management page and task logs.

8. Troubleshooting

SymptomCheck first
No 3D objectsThree.js loading, AgentGraph3D initialization, snapshot API success, and current account access to Agents/Groups
Unexpected pillar heighttaskStatusCounts, runningTaskCount, and filters; height is not a score
Paused without the magenta HIL styleThe task may be an ordinary pause, or another entry may already have resolved the pending request; check GetHumanRequests and the Host instance
No skill markersAgent bindings, MCP Endpoint, A2A publication, and snapshot fields
Labels remain crowdedLock one Group or filter by Group/status; confirm the browser loaded the versioned agent-3d.js asset
3D differs from task detail3D is a polled snapshot; task detail and SSE/history are the authoritative evidence for one task

Related pages:

  • AgentsManager HIL and Workflow integration
  • NeuChar Workflow
Edit this page on GitHub
Last Updated:
Contributors: JeffreySu