如何构建一个每天自动变聪明、无需额外投入的 N8N 和 Obsidian 业务脑

大多数企业主管理知识的方式和管理邮件一样。

被动反应。

信息来了。处理掉。继续前进。那些本可以让明天决策更明智的信息，就在处理今日任务的过程中流失了。

N8N 和 Obsidian 业务脑从根本上解决了这个问题。

Obsidian 成为你业务的永久记忆。每一个决策、每一个洞见、每一个模式、每一条你的业务产生的信息，都以纯文本文件的形式存储，永远不会消失，也永远不需要订阅来访问。

N8N 成为连接业务与这份记忆的神经系统。它监测新信息，自动处理，将其路由到知识库的准确位置，从中生成情报，并在你意识到需要之前就将其交付给你。

结果就是，一个在你每天运营中都会变得更聪明的业务脑——因为每一天的运营都会增加它积累的智慧。

而且，所有这些都不需要你在正常运营业务之外付出任何额外努力。

这是一份完整的构建指南。

架构

系统由三个协同工作的层级组成。

层级 1：知识层 — Obsidian

以一致的格式组织的纯文本 Markdown 文件。所有业务知识都永久存储在此。决策、会议记录、竞争情报、客户洞察、项目状态、财务观察。所有文件都由你完全拥有。

层级 2：自动化层 — N8N

按照计划运行或由事件触发的可视化工作流。N8N 监测你的业务工具获取新信息，通过 Claude 处理，并将输出直接写入你的 Obsidian 知识库。它还从知识库读取信息，以生成情报报告和简报。

层级 3：智能层 — Claude API

每个需要产生智能输出的 N8N 工作流都通过 Anthropic API 调用 Claude。Claude 读取原始业务数据，将其综合为结构化知识，识别模式，并生成需要真正推理而不仅仅是数据转换的输出。

搭建基础

安装 N8N：

与 Obsidian 集成最有效的设置是在与你的知识库同一台机器上本地运行 N8N。

npm install -g n8n
n8n start

N8N 运行在 localhost:5678。你的 Obsidian 知识库可以通过正常路径访问。N8N 中的所有文件系统节点都可以直接读写知识库文件，无需额外配置。

配置 Claude API 节点：

创建一个你将在所有工作流中复用的 HTTP Request 节点：

• URL: https://api.anthropic.com/v1/messages
• Method: POST
• Header: x-api-key: [YOUR API KEY]
• Header: anthropic-version: 2023-06-01
• Header: content-type: application/json

标准请求体：

{
  "model": "claude-opus-4-8",
  "max_tokens": 4096,
  "messages": [
    {
      "role": "user",
      "content": "{{ $json.prompt }}"
    }
  ]
}

响应提取：{{ $json.content[0].text }}

业务脑的知识库结构：

BUSINESS-BRAIN/
    00-INBOX/
        [raw N8N captures land here]
    
    01-INTELLIGENCE/
        customers/
            [customer insight notes]
        market/
            [competitive and market intelligence]
        decisions/
            [every significant business decision]
        patterns/
            [recurring patterns identified]
    
    02-OPERATIONS/
        projects/
            [active project folders]
        meetings/
            [meeting notes by date]
        reviews/
            [weekly and monthly reviews]
    
    03-FINANCIAL/
        [financial observations and patterns]
    
    04-PIPELINE/
        [sales pipeline notes]
    
    05-OUTPUTS/
        briefings/
            [daily and weekly briefings]
        syntheses/
            [topic syntheses]
        alerts/
            [pattern alerts and flags]
    
    06-SYSTEM/
        CLAUDE.md
        skills/

编写业务 CLAUDE.md：

# Business Brain — CLAUDE.md

## Business Identity
Company: [YOUR COMPANY NAME]
What we do: [ONE CLEAR PARAGRAPH]
Stage: [EARLY/GROWTH/ESTABLISHED]
Revenue model: [HOW YOU MAKE MONEY]
Primary customer: [SPECIFIC ICP]

## Current Strategic Priorities
1. [MOST IMPORTANT THING THIS QUARTER]
2. [SECOND MOST IMPORTANT]
3. [THIRD MOST IMPORTANT]

## Active Projects
[PROJECT NAME]: [STATUS] — Next: [SPECIFIC ACTION]
[Repeat for each active project]

## Active Theses
[BELIEF 1 ABOUT YOUR MARKET OR BUSINESS]
Evidence for: [WHAT SUPPORTS IT]
Counter-evidence: [WHAT CHALLENGES IT]

## Competitive Landscape
Primary competitors: [LIST WITH ONE-LINE DESCRIPTIONS]
Our differentiation: [SPECIFIC AND HONEST]

## Known Patterns
[PATTERNS IDENTIFIED FROM VAULT ANALYSIS]
[This section grows as the brain learns]

## Known Risks
[SPECIFIC RISKS WORTH MONITORING]

## Decision Log Context
When logging decisions always capture:
- The decision made
- The key assumption it rests on
- The expected outcome
- The review date

## Intelligence Standards
Strong signal: information that would change 
a decision or challenge an active thesis.

Weak signal: interesting but not immediately 
actionable or thesis-relevant.

Noise: everything else. Do not log.

## Output Instructions
All outputs save to 05-OUTPUTS with date prefix.
All decisions save to 01-INTELLIGENCE/decisions.
All customer insights to 01-INTELLIGENCE/customers.
All market intelligence to 01-INTELLIGENCE/market.

八个核心工作流

工作流 1：晨间商业简报

每天早上 6 点，这个工作流会读取你的整个业务脑，并生成一份结构化的简报，在一切事务争夺你注意力之前告诉你什么最重要。

N8N 结构：

Schedule Trigger (6AM daily)
    ↓
Read File Node → CLAUDE.md
    ↓
Read All Files → 02-OPERATIONS/projects/ (all .md files)
    ↓
Read File Node → Yesterday's meeting notes if exists
    ↓
HTTP Request → Brave Search (industry news relevant to business)
    ↓
Set Node (combine all content into prompt)
    ↓
HTTP Request → Claude API
    ↓
Write File Node → 05-OUTPUTS/briefings/[DATE]-morning-brief.md
    ↓
Telegram Node → notify brief is ready

Claude 提示词：

Read the attached business context and generate 
a morning brief for the business owner.

CLAUDE.md context: {{ $node["Read CLAUDE.md"].json.content }}

Active project notes: {{ $node["Read Projects"].json.content }}

Recent meeting notes: {{ $node["Read Meetings"].json.content }}

External industry news: {{ $node["Search News"].json.results }}

Generate the morning brief in this format:

---
# Morning Brief — [DATE]

## THE ONE THING
[Single most important thing for this business today.
Specific. Actionable. Based on actual vault context.]

## PROJECT PULSE
[For each active project: one line on status 
and the specific next action]

## OVERNIGHT INTELLIGENCE
[1-3 external developments relevant to the business.
Only include if genuinely signal not noise.]

## DECISION REQUIRED
[Any decision that has been sitting open for more 
than 7 days based on vault context]

## PATTERN FLAG
[Any pattern from recent notes worth flagging]
---

Be specific. Reference actual project names and 
actual next actions. Generic business advice is 
worthless. Vault-grounded specificity is everything.

工作流 2：会议智能处理器

每次你将会议转录或粗略笔记放入收件箱时，这个工作流会将其处理为结构化的智能信息。

触发器： 在 00-INBOX 中创建包含 “meeting” 名称的新文件

N8N 结构：

File Trigger → Watch 00-INBOX for new files
    ↓
IF Node → filename contains "meeting"
    ↓
Read Binary File → get file content
    ↓
Read File → CLAUDE.md
    ↓
HTTP Request → Claude API
    ↓
Write Binary File → 02-OPERATIONS/meetings/[DATE]-[title].md
    ↓
Trigger → Inbox Cleaner (move processed file to archive)

Claude 提示词：

Process this meeting transcript or notes into 
a structured intelligence note for the business brain.

CLAUDE.md context: {{ $node["Read CLAUDE.md"].json.content }}

Raw meeting content: {{ $node["Read Meeting"].json.content }}

Produce the meeting intelligence note in this format:

---
type: meeting
date: [EXTRACT FROM CONTENT OR USE TODAY]
attendees: [EXTRACT FROM CONTENT]
project: [WHICH PROJECT THIS RELATES TO]
---

# [MEETING TITLE]

## Decisions Made
[Specific decisions. Format each as:]
DECISION: [what was decided]
ASSUMPTION: [key belief this rests on]
OWNER: [who is responsible]
REVIEW: [when to check if correct]

## Action Items
[Specific actions with owners and deadlines]
- [ ] [ACTION] — [OWNER] — [DEADLINE]

## Intelligence Captured
[Any competitive intelligence, customer insights, 
or market information mentioned]

## Open Questions
[Unresolved questions that need follow-up]

## Connection to Strategy
[How this meeting connects to current strategic 
priorities in CLAUDE.md]

工作流 3：客户情报聚合器

此工作流监控你的 CRM、支持工单和客户沟通渠道，提取洞察并将其存入知识库的客户情报层。

N8N 结构：

Schedule Trigger (daily at 9PM)
    ↓
HTTP Request → CRM API (new deals/contacts today)
    ↓
HTTP Request → Support Tool API (tickets resolved today)
    ↓
HTTP Request → Email API (customer emails received today)
    ↓
Merge Node (combine all customer data)
    ↓
IF Node → any items to process?
    ↓
HTTP Request → Claude API
    ↓
Write Binary File → 01-INTELLIGENCE/customers/[DATE]-customer-intel.md

Claude 提示词：

Extract and structure customer intelligence from 
today's customer interactions.

CLAUDE.md context: {{ $node["Read CLAUDE.md"].json.content }}

CRM data: {{ $node["CRM"].json.data }}
Support tickets: {{ $node["Support"].json.tickets }}
Customer emails: {{ $node["Email"].json.messages }}

For each meaningful customer interaction produce:

INSIGHT TYPE: [Pain point / Feature request / 
              Competitive mention / Churn signal / 
              Expansion signal / Satisfaction signal]

CUSTOMER SEGMENT: [Which customer segment this represents]

CONTENT: [What was said or what happened in 2-3 sentences]

IMPLICATION: [What this means for the business in one sentence]

THESIS IMPACT: [Does this support or challenge any active 
thesis in CLAUDE.md? If yes name the thesis.]

Only extract genuine intelligence. Skip routine 
interactions with no signal value.

Group insights by type and produce a daily 
customer intelligence summary.

工作流 4：竞争情报监控器

此工作流每周一和周四运行，扫描竞争动态和市场变化。

N8N 结构：

Schedule Trigger (Monday and Thursday 7AM)
    ↓
Read File → CLAUDE.md (extract competitor names)
    ↓
Loop Over Competitors
    ↓
HTTP Request → Brave Search ([competitor] news)
    ↓
HTTP Request → Brave Search ([competitor] product updates)
    ↓
Merge Node → all competitive findings
    ↓
HTTP Request → Claude API
    ↓
Write Binary File → 01-INTELLIGENCE/market/[DATE]-competitive.md
    ↓
IF Node → any significant finding?
    ↓
Telegram Node → alert if significant

Claude 提示词：

Analyze these competitive intelligence findings 
and produce a structured update.

CLAUDE.md context including competitors: 
{{ $node["Read CLAUDE.md"].json.content }}

Competitive search results: 
{{ $node["Merge Competitive"].json.results }}

For each competitor produce:

STATUS: [Active development / Stagnant / 
         Strategic shift / Significant news]

FINDING: [What was discovered in 2-3 sentences]

SIGNIFICANCE: [High / Medium / Low for our business]

OUR RESPONSE: [Should we do anything about this?]

THESIS IMPACT: [Does this support or challenge any 
active thesis in CLAUDE.md?]

Flag any finding rated HIGH significance for 
immediate attention via the alert system.

If nothing significant happened for a competitor 
write "No significant developments" and move on.
Don't pad the report with noise.

工作流 5：决策记录与审核器

此工作流处理两个功能：当你将决策放入收件箱时记录决策，并提醒已过审核日期的决策。

Part A: 决策记录器

File Trigger → Watch 00-INBOX for files containing "decision"
    ↓
Read Binary File → get raw decision content
    ↓
HTTP Request → Claude API (structure the decision)
    ↓
Write Binary File → 01-INTELLIGENCE/decisions/[DATE]-[topic].md
    ↓
Archive original file

Part B: 决策审核器

Schedule Trigger (every Monday 8AM)
    ↓
Read All Files → 01-INTELLIGENCE/decisions/
    ↓
Code Node → filter files where review_date < today
    ↓
IF Node → any decisions due for review?
    ↓
HTTP Request → Claude API (generate review prompts)
    ↓
Write Binary File → 05-OUTPUTS/alerts/[DATE]-decision-reviews.md
    ↓
Telegram Node → notify decisions need review

决策结构化提示词：

Structure this decision for the business brain.

Raw decision content: {{ $json.content }}

Business context from CLAUDE.md: {{ $node["CLAUDE.md"].json.content }}

Produce the decision note:

---
type: decision
date: [TODAY]
status: active
review_date: [SUGGEST BASED ON DECISION TYPE]
---

# Decision: [DECISION TOPIC]

## The Decision
[Clear statement of what was decided]

## Context
[Why this decision needed to be made]

## Alternatives Considered
[What else was on the table]

## Key Assumption
[The single most important belief this decision rests on]

## Expected Outcome
[What success looks like]

## Review Criteria
[How you will know if this decision was right]

## Risk
[What could make this decision wrong]

工作流 6：每周业务回顾

每周日晚上 7 点，此工作流读取整周的知识库活动，并生成一份全面的业务回顾。

N8N 结构：

Schedule Trigger (Sunday 7PM)
    ↓
Read All Files → 02-OPERATIONS/meetings/ (this week)
    ↓
Read All Files → 01-INTELLIGENCE/customers/ (this week)
    ↓
Read All Files → 01-INTELLIGENCE/market/ (this week)
    ↓
Read All Files → 05-OUTPUTS/briefings/ (this week)
    ↓
Read File → CLAUDE.md
    ↓
HTTP Request → Claude API
    ↓
Write Binary File → 02-OPERATIONS/reviews/[DATE]-weekly.md
    ↓
HTTP Request → Claude API (update CLAUDE.md recommendations)
    ↓
Telegram Node → weekly review ready

Claude 提示词：

Generate a comprehensive weekly business review 
from this week's vault activity.

CLAUDE.md: {{ $node["CLAUDE.md"].json.content }}
Meetings this week: {{ $node["Meetings"].json.content }}
Customer intelligence: {{ $node["Customer Intel"].json.content }}
Market intelligence: {{ $node["Market Intel"].json.content }}
Daily briefs: {{ $node["Briefs"].json.content }}

Produce the weekly review:

---
# Weekly Business Review — [WEEK OF DATE]

## THE WEEK IN ONE LINE
[Single honest sentence about what this week was]

## WHAT MOVED
[Specific progress on strategic priorities.
Reference actual projects and outcomes.]

## WHAT STALLED
[Honest assessment of what didn't move.
Name the specific reason for each.]

## CUSTOMER INTELLIGENCE SUMMARY
[What did customers tell us this week.
Any patterns worth noting.]

## COMPETITIVE DEVELOPMENTS
[Any competitive intelligence from the week 
worth carrying forward.]

## DECISION QUALITY CHECK
[Were any decisions made this week based on 
assumptions that look shakier after the week's data?]

## PATTERN OF THE WEEK
[One theme that appeared across multiple 
different inputs this week]

## NEXT WEEK'S SINGLE PRIORITY
[The one thing that matters most next week 
based on this week's evidence. Specific.]

## RECOMMENDED CLAUDE.MD UPDATES
[Any changes to strategic priorities, active 
theses, or known patterns that this week's 
data supports]
---

工作流 7：模式检测器

此工作流每月运行，识别重复出现的模式