@QingQ77: 把公司文档自动整理成知识 Wiki,通过 MCP 让每个员工的 AI 客户端拿到对口的上下文,不用再手动粘贴。 https://github.com/nduckmink/arkon Arkon 是可自部署的企业 AI 知识中枢。上传 SO…
摘要
Arkon 是一款可自部署的企业 AI 知识中枢,能将公司文档自动编译成交叉链接的知识 Wiki,并通过 MCP 协议让员工的 AI 客户端(如 Claude Desktop)按权限自动获取相关上下文,无需手动粘贴文档。
查看缓存全文
缓存时间: 2026/05/09 05:44
把公司文档自动整理成知识 Wiki,通过 MCP 让每个员工的 AI 客户端拿到对口的上下文,不用再手动粘贴。 https://github.com/nduckmink/arkon Arkon 是可自部署的企业 AI 知识中枢。上传 SOP、政策、产品文档后,LLM Agent 把它们编译成交叉链接的 Wiki。员工用 MCP Token 连上 Claude Desktop 等客户端,按权限自动拿到相关知识、原文和 AI Skills。
nduckmink/arkon
Source: https://github.com/nduckmink/arkon

Arkon - Enterprise AI Knowledge Hub
Connect your organization’s knowledge to any Ai Client. Self-hosted, on-premise.
Arkon is the central layer between your documents and your employees’ AI clients. Upload your SOPs, policies, product specs, and internal docs - Arkon compiles them into a structured knowledge wiki and serves it directly to Claude via MCP. Every employee gets the right context, automatically, without copy-pasting.
Setup Guide · Architecture · Wiki System · MCP & Claude · AI Skills · Access Control
The problem
Most organizations adopt AI team-by-team, with no shared knowledge, inconsistent context, and no visibility into what information Ai Client is actually working with. Every employee manually pastes documents, repeats the same background, and gets different answers depending on what they remembered to include.

Arkon treats your AI client as a managed organizational resource - not a personal chatbot.
How it works

Knowledge compounds. Every document you add enriches the existing wiki rather than creating isolated fragments. By the time an employee asks Claude a question, the answer has already been synthesized from dozens of sources.
Features
Knowledge Wiki
Documents are compiled into a persistent, interlinked wiki by an LLM agent - not just indexed. Each page covers a specific entity, concept, or topic. Pages cross-reference each other. The wiki grows smarter as more documents are added.
- Three-panel wiki browser: page tree, content, backlinks & outlinks
- Full-text and semantic search
- Knowledge graph visualization
- Organize by knowledge type (SOP, Product, HR Policy, etc.)
- Version history and rollback for every page
- Draft proposal → editor review → approval workflow
Workspaces
Cross-functional knowledge contexts for projects, clients, or initiatives.
Create a workspace → add members from any department → attach documents. Each workspace has its own scoped wiki, document list, and member roster. Members see their workspace knowledge automatically through Claude.
- Role-based membership: Viewer, Contributor, Editor, Admin
- Scoped wiki and document management
- Contributors propose wiki edits; editors review and approve
AI Skills
Upload custom agent packages and make them available to employees through Claude. Skills are versioned, department-scoped, and distributed via MCP.
MCP Server
Employees connect Claude Desktop (or any MCP client) to Arkon using a personal token. Claude gets access to the compiled wiki, raw source documents, and AI skills - all filtered to the employee’s permission scope.
→ See MCP & Claude for the full tool reference.
Access Control
Fine-grained RBAC at department level plus workspace membership roles. Admins define roles with granular permissions; employees inherit access based on department or explicit assignment.
→ See Access Control for the full permission model.
Quick Start (Docker)
Prerequisites: Docker, Docker Compose, an AI provider API key (Google, OpenAI, or Anthropic).
git clone https://github.com/nduckmink/arkon.git
cd arkon
cp .env.docker.example .env.docker
Edit .env.docker - set at minimum:
SECRET_KEY=<run: python -c "import secrets; print(secrets.token_urlsafe(32))">
[email protected]
DEFAULT_ADMIN_PASSWORD=your-secure-password
docker compose --env-file .env.docker up -d --build
Open http://localhost:3119 and log in with your admin credentials.
Go to Settings → configure your embedding model, LLM, and (optionally) vision model. Then upload your first document from Knowledge Base.
→ See Setup Guide for the full walkthrough including development mode.
Connecting Claude
Once an employee account is created and an MCP token is generated:
{
"mcpServers": {
"arkon": {
"url": "https://your-arkon-server/mcp",
"headers": {
"Authorization": "Bearer ark_xxxxxxxxxxxx"
}
}
}
}
Add this to claude_desktop_config.json and restart Claude Desktop. The employee’s compiled knowledge is immediately available.
→ See MCP & Claude for the complete setup and tool reference.
Architecture

Stack: FastAPI · PostgreSQL + pgvector · Redis (arq) · MinIO · Next.js · Tailwind CSS
AI providers (your choice): Google · OpenAI · Anthropic · Ollama · Voyage · Cohere
Outbound network: configured AI provider only. No telemetry, no external calls.
→ See Architecture for the full technical breakdown.
Roadmap
- LLM Wiki Agent - documents compiled into persistent, interlinked wiki
- Wiki browser - three-panel layout with graph visualization
- MCP Server with scoped knowledge access
- Ingestion pipeline - PDF, DOCX, DOC, URLs, images with vision captions
- Workspaces - scoped wiki, documents, and members
- Wiki draft & revision system - propose, review, approve, rollback
- AI Skills - versioned, department-scoped agent packages
- Full RBAC - department permissions + workspace membership roles
- Audit log
- Regex Search for Human and AI
- Arkon CLI - one-command employee setup
- Notification system for draft review requests
- Usage analytics dashboard
License
Arkon is licensed under the PolyForm Noncommercial License 1.0.0.
Free to use for internal tooling, research, personal projects, and non-profit use.
Need a commercial license or custom integration? We help organizations integrate Claude, custom AI agents, and MCP servers into their existing infrastructure - from connecting to internal databases and legacy systems to building purpose-built agents for specific business processes.
相似文章
@MindOS_Lisa: https://x.com/MindOS_Lisa/status/2052766937931973065
作者介绍了基于Karpathy知识库逻辑,结合Obsidian、Claude Code和LLM Wiki搭建个人知识管理系统的方法,实现内容自动收录、卡片生成与文章输出的完整闭环。文章区分了内容、信息与知识三个层次,并提供了具体搭建步骤。
Full Claude Cowork Tutorial: The AI That Replaces Your Assistant
本文详细介绍了 Claude Co-work 功能,这是一个驻留在本地计算机上的 AI 智能体,能够直接访问文件、并行处理任务并与日历和邮件等应用集成,从而完成复杂的多步骤工作流。文章通过整理文件和准备收购会议等实战案例,展示了其在自动化办公和战略规划方面的强大能力。
在 ChatGPT 中利用公司知识更聪明地工作
OpenAI 为 ChatGPT Business、Enterprise 和 Edu 推出了公司知识功能,使 AI 能够访问和综合来自 Slack、SharePoint、Google Drive 和 GitHub 等连接的工作应用中的信息,提供带有引用的情境化业务答案。
@axichuhai: OpenAI 联合创始人 Karpathy 的知识库里塞着 40 万字、上百篇文章,但他自己从来不整理一个字。 前几天他把方法公开在 GitHub 上,一条推文冲上 2000 w 曝光。 很多人收藏夹几百上千条,两周后连放哪儿都忘了。 收…
Andrej Karpathy open-sourced his personal knowledge-management workflow on GitHub, revealing how he organizes a 400k-word archive without manual curation.
@indigox: 强烈推荐 Markdown 专用编辑器 cogito.md!简洁优雅快速,所有项目可以按文件夹组织在左侧导航,能在文件和项目维度调用 Claude Code or Codex 作为 Agent 服务集成,可视化搭建知识库的利器!比 Obs…
cogito.md 是一款简洁优雅的 Markdown 专用编辑器,支持文件夹组织项目,可集成 Claude Code 或 Codex 作为 Agent 服务,适合可视化构建知识库,被认为比 Obsidian 更适合 Agent 工作流。