BossAgents × WeChat Intelligent Agent Ecosystem: Learning, Integration, and Evolution Plan

Version: v1.0 | Date: 2026-06-24 | Based on the latest 2025-2026 WeChat/Tencent AI Agent ecosystem research

1. Current State Benchmarking Analysis

1.1 BossAgents Current Capability Matrix vs. WeChat Ecosystem

II. Option One: Learning from WeChat Intelligent Agents — Six Major Upgrade Directions

2.1 ? P0 Priority: Memory System Upgrade (Benchmarking Hy-Memory)

Current Issues:

• BossAgents only has MEMORY.md (plain text) and audit logs (JSONL sequential files)
• No structured memory layering, no evolution tracking, no active learning
• Digital employees "start fresh every day," lacking continuous understanding of the boss/enterprise

Hy-Memory Reusable Architecture:

┌─────────────────────────────────────────────────────────────┐
│              BossAgents 记忆系统升级方案                      │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  Layer 1: Atomic Traces (原子痕迹)                           │
│  └─ 每次 Feishu/Web 交互的原始消息 + 执行结果                │
│     → 复用现有 audit-logs/*.jsonl                            │
│                                                             │
│  Layer 2: Atomic Facts (原子事实)                            │
│  └─ 从交互中提取事实：老板偏好、供应商偏好、常用操作          │
│     → 新增 memory/facts.json 结构化存储                      │
│                                                             │
│  Layer 3: Identity Profile (用户画像) ★ 核心新增             │
│  └─ 企业信息、老板角色、常用对象类型、审批偏好                │
│     → 新增 memory/profile.json + 企业画像自动更新            │
│                                                             │
│  Layer 4: Session Summaries (会话摘要)                       │
│  └─ 已通过 LCM (Lossless Context Management) 部分覆盖        │
│     → 增强：按"业务主题"聚合（BOM/采购/变更）                │
│                                                             │
│  Layer 5: Mental Models (心智模型)                            │
│  └─ 深度理解老板决策模式：风险偏好、成本敏感度、创新意愿      │
│     → 新增 memory/mental-models.json，由 System 2 异步构建   │
│                                                             │
│  Layer 6: Forward-looking Intentions (前瞻意图)               │
│  └─ 预判老板下一步需求："上周优化了BOM成本，这周该看库存了"   │
│     → 新增 memory/intentions.json，心跳任务定期更新           │
│                                                             │
│  System 1 (Day Shift): 实时写入                               │
│  └─ CapabilityDispatcher.execute() 每次执行后提取关键事实     │
│                                                             │
│  System 2 (Night Shift): 异步分析                             │
│  └─ 定时 cron job，分析 audit logs → 更新 Mental Models       │
│                                                             │
│  演化链 (Evolution Chain):                                    │
│  └─ supersedes 指针链接旧/新记忆，保留决策演化历史             │
│                                                             │
└─────────────────────────────────────────────────────────────┘

具体实施方案：

| Step | Content | Effort Estimate |

|------|------|-----------|

| 1 | Create new server/services/memory-service.js | 3 days |

| 2 | Layer 1-3 Implementation: Trace → Fact → Profile extraction chain | 5 days |

| 3 | System 1 Integration: Automatic fact extraction after Dispatcher execution | 2 days |

| 4 | System 2 cron job: Nightly analysis + profile update | 3 days |

| 5 | Layer 5-6 Implementation: Mental model + prospective intent | 5 days |

| 6 | Evolution chain: supersedes pointer + version management | 3 days |

2.2 ? P1 Priority: Multi-Agent parallel execution (benchmarked against WorkBuddy + ADP 3.0)

Current Issues:

• LiteScheduler serial scheduling: One employee per cron, mutex lock prevents concurrency
• WorkBuddy has already implemented multi-agent parallel processing of different subtasks

Upgrade Plan:

现有架构:
  cron(15min) → Staff A → runOnce → _runWorker → 完成
  cron(30min) → Staff B → runOnce → _runWorker → 完成
  (串行，独立 cron)

升级后:
  老板: "比较三家供应商报价并生成采购建议"
       │
       ▼
  CapabilityDispatcher
       │
       ▼
  TaskDecomposer (新增)  ← LLM 拆解
       │
       ├──→ SubTask 1: 查询供应商A报价  ──→ Staff A (并行)
       ├──→ SubTask 2: 查询供应商B报价  ──→ Staff B (并行)
       ├──→ SubTask 3: 查询供应商C报价  ──→ Staff C (并行)
       │
       ▼
  ResultAggregator (新增)  ← LLM 汇总
       │
       ▼
  返回："推荐供应商B，综合评分最高"

新增模块：

• server/core/task-decomposer.js — Task decomposition + parallel scheduling
• server/core/result-aggregator.js — Result aggregation + conflict resolution
• Add runParallel(staffIds[], subTasks[]) method to LiteScheduler

2.3 ? P1 Priority: Enhance LLM Router (benchmarked against Mixed Yuan model matrix)

Current State: MTCLAW Worker/Solver two-way routing

Upgrade: Three-layer model routing + integration of domestic Mixed Yuan models

请求进入
    │
    ▼
  Layer 1: 规则引擎（0ms，0 cost）
    │ 命中? → 直接执行
    ▼
  Layer 2: 端侧小模型 Worker（~70ms）
    │ 高频查询/校验/列表 → 本地 Gemma/Hunyuan-1.8B
    ▼
  Layer 3: 云端大模型 Solver（~500ms）
    │ 复杂推理/Create/分析 → DeepSeek/Hunyuan-2.0
    ▼
  Fallback: 降级策略（四级）

新增混元模型接入点：

• HUNYUAN_API_KEY environment variable → Tencent Cloud API
• Model selection: hunyuan-2.0-thinking-20251109 for complex reasoning
• Cost consideration: Hunyuan's domestic pricing may be better than OpenRouter

2.4 ? P2 Priority: Knowledge Base Agentic RAG (benchmarked against ima.copilot + ADP 3.0)

Current state: SCIOT rule engine handles deterministic matching, LLM is used when no match is found

Upgrade: Learn from ADP 3.0 "Agentic RAG" — Agent actively evaluates retrieval quality

现有: 规则引擎命中? → 执行 : LLM
升级: 规则引擎 → 命中? → 执行
                  ↓ 未命中
              RAG 检索 SCIOT Knowledge base
                  ↓
              Agent 自评检索质量
                  ↓ 不足则重新检索 / 换策略
              → LLM 基于检索结果推理

2.5 ? P2 优先级：用户画像系统（对标 ima.copilot 记忆）

ima.copilot 的记忆系统四模块：

1. Copilot setup → BossAgents' SOUL.md / IDENTITY.md
2. User profile → New: memory/user-profile.json, learns automatically from interactions
3. Long-term memory → Upgrade MEMORY.md to structured memory
4. Experience & techniques → New: memory/lessons.json, accumulates digital employee execution experience

2.6 ? P3 Priority: Self-evolution closed-loop enhancement (benchmarked against evolution chain)

Current state: sciot_evolution_logs + FeedbackEngine analysis

Upgrade: Add time dimension to the evolution chain

当前: 修正记录 → 分析 → 生成改进建议
升级: 修正记录 → 版本链 → 趋势分析 → 预测性建议
      每次修正不是覆盖，而是追加新版本
      supersedes 指针保留完整演化历史

三、方案二：与微信智能体结合 — 四大集成路径

3.1 ? P0 优先级：接入 ClawBot — 开通微信原生入口

这是最高价值、最紧迫的集成点。

BossAgents 后端 (现有)
    │
    ├── Web 入口 (现有)
    ├── Feishu 入口 (现有)
    ├── Mini Program 入口 (bossagents-miniapp, 开发中)
    │
    └── ★ WeChat ClawBot 入口 (新增) ← iLink 协议
         │
         ▼
    ilinkai.weixin.qq.com
         │
         ▼
    老板微信 → 右滑 → ClawBot → @BossAgents
    "Optimize BOM-123 成本降低 10%"
         │
         ▼
    OpenClaw Gateway → CapabilityDispatcher
         │
         ▼
    数字员工执行 → 结果回微信

接入步骤：

1. Install: npx -y @tencent-weixin/openclaw-weixin-cli@latest install
2. WeChat side: Me → Settings → Plugins → ClawBot → Enable → Scan QR code to pair
3. BossAgents configuration OpenClaw channel: wechat channel
4. CapabilityDispatcher already has source: 'wechat' route support
5. Test: Send command within WeChat → Digital employee executes → Result returns to WeChat

Effort estimate: 1-2 days (main work in configuration and testing)

3.2 ? P1 Priority: bossagents-miniapp access to WeChat AI ecosystem

Current status: bossagents-miniapp is a uni-app project with basic architecture already in place

Opportunity: WeChat opens mini-program AI ecosystem access on June 8, 2026

Two modes:

A. Automatic mode (quick win):

• Authorize WeChat to read mini-program source code upon audit submission
• AI automatically analyzes page structure and functionality
• Users can operate the mini-program via natural language within WeChat AI
• Zero additional development

B. Development mode (deep integration):

• Build custom integration based on WeChat Mini Program AI SDK
• Embed "AI Brain" conversation entry within mini-program (design already exists)
• Call WeChat speech recognition → backend CapabilityDispatcher

Recommendation: Dual-mode + progressive roadmap

Phase 1 (1周): 自动模式 → 快速上线，验证流量
Phase 2 (2周): 开发模式 → 深度集成，差异化体验

3.3 ? P1 优先级：接入混元模型（国产合规 + 成本优化）

价值：

• Enterprise customers have compliance requirements for domestic large models
• Hunyuan pricing may be more favorable than OpenRouter
• Hunyuan 2.0 specifically enhances Agent capabilities
• Hunyuan 3.0 will enhance long-term memory and multi-turn dialogue

Access solution:

// server/digital-staff/llm-brain.js 新增 provider
providers: {
  openrouter: { ... },
  hunyuan: {
    baseURL: 'https://api.hunyuan.cloud.tencent.com/v1',
    apiKey: process.env.HUNYUAN_API_KEY,
    models: {
      fast: 'hunyuan-2.0-instruct-20251109',    // 简单任务
      think: 'hunyuan-2.0-thinking-20251109',   // 复杂推理
    }
  }
}

3.4 ? P2 优先级：接入 Hy-Memory 插件

现状： Hy-Memory 已发布为 OpenClaw 插件，一行命令安装

Value: Directly obtain six-layer memory + dual-system + evolution chain capabilities

Access method:

# OpenClaw skill install
npx openclaw skill install hy-memory

注意： Hy-Memory 是为 OpenClaw 设计的。BossAgents 如果直接运行在 OpenClaw 上则零成本接入；如果独立运行，需要适配层。

IV. Option 3: Learn from WorkBuddy's best practices

4.1 "No deployment · Install and use" — Lower BossAgents deployment threshold

Current status: BossAgents requires Node.js + SQLite + manual .env configuration + DB initialization

Benchmark: WorkBuddy install-and-use, zero configuration

Improvement plan:

• One-click startup script: Enhance start.bat → auto-detect environment → auto-install dependencies → auto-initialize DB → launch
• Docker Image: Already have docker-compose.yml, but can be optimized to a single docker run command
• Windows Installer: Package as .exe using pkg/nexe (boss frequently uses Windows)
• Configuration Wizard: Guide users to fill in necessary information like API Key on first launch

4.2 "Send tasks on mobile, execute automatically on PC" — Remote Control Enhancement

WorkBuddy Highlight: Send message via WeChat/WeCom → WorkBuddy on PC automatically executes → Results pushed back to phone

BossAgents Benchmark:

• Already implemented: Feishu → Dispatcher → Digital Employee → Results flow back to Feishu
• Enhancement: After integrating ClawBot → Send command via WeChat → Server executes → Results return to WeChat
• Enhancement: WebSocket real-time push of execution progress (WebSocket already exists in mini-program design plan)

4.3 OpenClaw Skills Compatibility — Reuse Ecosystem

Current Status: BossAgents runs on OpenClaw, natively compatible with Skills ecosystem

Recommendation:

• Package BossAgents core capabilities as an OpenClaw Skill
• Digital employee configuration → Skill configuration
• Rule engine → Tool inside Skill
• Publish to SkillHub community

4.4 "Agent Suite" — From Super Individual to Super Team

WorkBuddy Enterprise Concept: Personal Agent → Team Agent Suite

BossAgents Benchmark:

• Existing: 11 digital employees (Procurement/BOM/ECR/Supplier/Cost/Content...)
• Enhancement: Team collaboration chain (one employee triggers the next)
• Enhancement: Role-based permissions (boss/manager/engineer see different employees)

4.5 ima.copilot Knowledge Base Mode — Enhancing SCIOT Experience

ima Highlight: "The Thinking Knowledge Base" — Integrated Search-Read-Write

BossAgents SCIOT Benchmark:

• Existing: 13,000+ industrial object definitions, 2,300+ rules
• Enhancement: "SCIOT Knowledge Base Dialogue" → Boss explores industrial knowledge using natural language
• Enhancement: Automatic sync from SCSAI → ima-style automatic updates

V. Priority Implementation Roadmap

Phase 1 — Quick Wins (1-2 weeks)

| # | Task | Type | Value | Effort |

|---|------|------|------|--------|

| 1 | Integrate ClawBot, enable native WeChat entry | Integration | ? Very High | 1-2 days |

| 2 | Connect bossagents-miniapp auto mode to WeChat AI | Integration | ? High | 0.5 days |

| 3 | Implement Memory System Layer 1-3 | Learning | ? Very High | 5 days |

| 4 | Optimize one-click deployment package | Learning | ? Medium | 2 days |

Phase 1 Milestone: Boss can directly command digital employees in WeChat, system begins remembering boss preferences

Phase 2 — Deep Enhancement (3-4 weeks)

| # | Task | Type | Value | Effort |

|---|------|------|------|--------|

| 5 | Memory System Layer 4-6 + System 2 | Learning | ? Very High | 8 days |

| 6 | Multi-Agent Parallel Execution | Learning | ? High | 5 days |

| 7 | Integrate Hunyuan Large Model | Integration | ? High | 3 days |

| 8 | Deep Integration of bossagents-miniapp Development Mode | Integration | ? High | 10 days |

| 9 | Enhanced LLM Router Three-Tier Architecture | Learning | ? Medium | 3 days |

Phase 2 Milestone: System possesses long-term memory + parallel execution + domestic model, mini-program deep experience

Phase 3 — Ecosystem Expansion (5-8 weeks)

VI. Architecture Evolution Panorama

                         ┌──────────────────────────┐
                         │     WeChat / 企微 / 元宝    │
                         │   (新增微信 ClawBot 入口) │
                         └────────────┬─────────────┘
                                      │ iLink
┌─────────────────────────────────────┼──────────────────────────────────┐
│                             OpenClaw Gateway                           │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────────────┐      │
│  │ WeChat   │  │ Feishu   │  │ MiniApp  │  │ Web Dashboard    │      │
│  │ Channel  │  │ Channel  │  │ Channel  │  │ Channel          │      │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘  └───────┬──────────┘      │
│       └──────────────┼────────────┼─────────────────┘                 │
│                      ▼                                                 │
│  ┌──────────────────────────────────────────────────────────────┐    │
│  │              CapabilityDispatcher (统一调度)                   │    │
│  │  ┌──────────┐  ┌──────────┐  ┌──────────┐                   │    │
│  │  │ 参数归一化│  │ 审计日志  │  │ 权限校验  │                   │    │
│  │  └──────────┘  └──────────┘  └──────────┘                   │    │
│  └──────────────────────────┬───────────────────────────────────┘    │
│                             ▼                                         │
│  ┌──────────────────────────────────────────────────────────────┐    │
│  │                   TaskDecomposer (★新增)                      │    │
│  │        LLM 拆解复杂任务 → 并行调度多员工                       │    │
│  └──────────────────────────┬───────────────────────────────────┘    │
│                             ▼                                         │
│  ┌──────────────────────────────────────────────────────────────┐    │
│  │              三层模型路由 (★增强)                              │    │
│  │  Layer 1: 规则引擎  Layer 2: Worker  Layer 3: Solver         │    │
│  │  (0ms, 0 cost)      (~70ms)           (~500ms)                │    │
│  │  新增混元模型接入                                              │    │
│  └──────────────────────────┬───────────────────────────────────┘    │
│                             ▼                                         │
│  ┌──────────────────────────────────────────────────────────────┐    │
│  │              CapabilityRuntime (六大能力)                      │    │
│  │  identify │ create │ repair │ optimize │ compare │ generate   │    │
│  │                    ★ 新增 Agentic RAG                         │    │
│  └──────────────────────────┬───────────────────────────────────┘    │
│                             ▼                                         │
│  ┌──────────────────────────────────────────────────────────────┐    │
│  │                   Memory Service (★新增)                       │    │
│  │  Traces → Facts → Profile → Summaries → Models → Intentions  │    │
│  │  System 1 (实时写入)  +  System 2 (异步分析)                  │    │
│  └──────────────────────────────────────────────────────────────┘    │
└──────────────────────────────────────────────────────────────────────┘

七、总结

核心结论

1. 最大的学习机会：记忆系统。 Hy-Memory 的六层记忆+双系统+演化链是 BossAgents 最需要学习的架构，当前 BossAgents 在这块几乎空白。

1. 最紧迫的结合点：WeChat ClawBot 入口。 这是"零成本、极高价值"的集成，1-2天即可让老板在微信里指挥数字员工。

1. 最值得借鉴的实践：WorkBuddy 的"免部署"哲学和"IM远程控制"模式。 BossAgents 的飞书交互已经走在正确的路上，接下来是微信通道和部署体验优化。

1. 最大的生态红利：小程序AI生态接入。 2026年6月刚开放，BossAgents 小程序正好赶上窗口期。

1. Long-term 方向：从"工具"到"伙伴"。 ima.copilot 从工具进化到 copilot 的路径，也是 BossAgents 从"数字员工平台"进化到"老板的AI伙伴"的方向。

一句话总结

BossAgents 要做的是：用 Hy-Memory 级别的记忆理解老板，通过微信 ClawBot 随时待命，借混元模型降本合规，以 WorkBuddy 式的零门槛体验，让工业数字员工真正成为老板口袋里的 AI 伙伴。