Intelligence Brain — AI-Native Company Intelligence Engine

Not a database. An engine.

Existing "brain" skills store facts. They cache. They retrieve. That's useful — but it's a library, not a brain. A real brain doesn't just store information. It ingests, digests, connects, infers, and outputs actionable intelligence.

Intelligence Brain is the engine that powered a live 7-department AI company's DataCenter — the department that gathered intelligence, managed agents, built neural chains, and produced cross-file inferences that no single document contained.

What This Is

An intelligence processing methodology with concrete implementation patterns. The brain:

1. Ingests raw information from multiple sources
2. Processes it through an 8-step metabolic pipeline
3. Infers cross-document insights via WISDOM_GRAPH and ACTION_PIPELINE
4. Routes outputs based on information classification
5. Manages agent lifecycles and inter-agent neural chains

The 8-Step Metabolic Pipeline

Raw information enters the company daily. Without a processing system, it accumulates as noise. The pipeline transforms noise into intelligence:

Step 1: READ     → Ingest incoming data (reports, research, outputs)
Step 2: PARSE    → Extract structured information from raw text
Step 3: DECOMPOSE→ Break into atomic facts, claims, and data points
Step 4: ORGANIZE → Categorize by domain, priority, and classification
Step 5: MERGE    → Cross-reference with existing knowledge, detect duplicates
Step 6: USE      → Apply to active decisions, content, or strategy
Step 7: DELETE   → Remove processed raw data (prevent accumulation rot)
Step 8: EXTRAPOLATE → Generate cross-file inferences (the "One More Thing" step)

Time constraint: All 8 steps must complete within 24 hours of ingestion. Unprocessed data decays in value exponentially.

Step Detail: DECOMPOSE

Input: "AI应用加速落地，2026年商业化提速，企业级应用增长300%"
Output:
  - [Claim] AI应用落地速度正在加快（来源：行业报告）
  - [Data] 企业级应用增长300%（需验证：来源、时间范围）
  - [Trend] 商业化从实验阶段进入加速阶段
  - [Action] 品牌部：可转化为"AI商业化"内容选题

Step Detail: EXTRAPOLATE

The most valuable step. Cross-file inference finds insights that exist between documents, not within them:

File A (品牌部报告): "小红书内容互动率下降15%"
File B (销售部报告): "知识星球付费转化率提升22%"

Cross-file inference: 免费平台引流效率下降 + 付费阵地转化上升
→ Action: 增加知识星球内容密度，减少小红书发布频率
→ Update: 变现漏斗权重调整

Cross-File Inference System

WISDOM_GRAPH

A network of insights connected across departmental outputs. Nodes are key findings. Edges are inferred relationships.

[小红书互动率↓] ──causes──→ [免费引流效率↓]
                                     │
                              ──suggests──→ [增加付费阵地投入]
                                     │
[知识星球转化↑] ──validates──→ [付费意愿存在]

ACTION_PIPELINE

Inferences that become actionable tasks, prioritized by urgency:

Max 4 items per priority tier. Forces ruthless prioritization.

ONE_MORE_THING

The highest-value inference: insights that connect domains nobody else thought to connect. Example from production:

Marketing data + Legal compliance data → "平台政策变化将影响内容分发策略"
Sales data + Finance data → "客户获取成本上升但客单价未变 → 定价策略需调整"

Neural Chain Architecture

The brain doesn't just process — it builds and maintains the communication infrastructure between agents.

Design Principles

1. Redundancy ≥ 3 — At least 3 parallel communication channels
2. No single point of failure — Any one channel can fail without breaking the system
3. Async-first — Agents read when ready, not when messaged
4. Auditability — Every inter-agent communication leaves a trace

Chain Topology

         ┌──────────┐
         │   CEO    │ (Decision only)
         └────┬─────┘
              │
    ┌─────────┼─────────┐
    │         │         │
┌───▼──┐ ┌───▼──┐ ┌───▼──┐
│Brand │ │Sales │ │Legal │
└───┬──┘ └───┬──┘ └───┬──┘
    │         │         │
    └─────────┼─────────┘
              │
    ┌─────────▼─────────┐
    │    DataCenter     │ (Hub)
    └─────────┬─────────┘
              │
    ┌─────────┼─────────┐
    │         │         │
┌───▼──┐ ┌───▼──┐ ┌───▼──┐
│Financ│ │Insp. │ │Admin │
└──────┘ └──────┘ └──────┘

DataCenter is the hub, but NOT a bottleneck. Messages route through it, not queue at it.

Information Classification & Model Routing

Every piece of information entering the brain gets classified, and classification determines which AI model processes it:

This is not just security — it's cost optimization. Why burn premium tokens on internal memos?

Agent Lifecycle Management

The brain tracks every agent in the organization:

Agent: Brand Department
Status: ACTIVE
Cron Jobs: 3 (daily learning, daily report, content publish)
Last Activity: 2026-06-25 15:52 CST
Consecutive Errors: 0
Model: deepseek/deepseek-v4-pro (publish), zai/glm-4-flash (learning)

When an agent goes silent (no activity for 24h) or errors spike (2+ consecutive failures), the brain flags it for inspection.

Implementation Guide

Phase 1: Pipeline Setup

1. Establish the company/data/ directory structure
2. Define data sources (department outputs, external feeds, research)
3. Implement the 8-step pipeline (initially manual, then cron-automated)
4. Set the 24h processing SLA

Phase 2: Inference Engine

1. Build WISDOM_GRAPH from first week of processed data
2. Generate initial ACTION_PIPELINE
3. Validate inferences against real outcomes
4. Tune the extrapolation heuristics

Phase 3: Neural Chain

1. Map existing communication paths
2. Identify single points of failure
3. Add redundant channels (minimum 3)
4. Test failure scenarios (kill one channel, verify others work)

Dependencies

Designed for the OpenClaw agent runtime:

• cron for scheduled pipeline execution
• sessions_send for neural chain messaging
• Workspace file system for data persistence
• Multi-model routing for cost optimization

Methodology is platform-agnostic. Implementation details are OpenClaw-specific.

Companion Skills

• opc-os-core — The complete one-person company operating system that this brain powers
• agent-org-manager — Multi-department setup and management
• cron-health-monitor — Cron job health monitoring (brain's self-diagnostic tool)

License

MIT — Free to use, modify, distribute. Built from production intelligence operations, not theory.