Skip to content

docs: Add FAQ section #82

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
meichuanyi wants to merge 2 commits into
base: master
Choose a base branch
from
Open

docs: Add FAQ section #82

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a05ef1f
docs: add FAQ section for common questions
meichuanyi May 2, 2026
17aa88b
docs: Add FAQ section
meichuanyi May 20, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
298 changes: 113 additions & 185 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,210 +1,138 @@
<div align="left">
<a href="./README.md"><img alt="README in English" src="https://img.shields.io/badge/English-d9d9d9" height="20"></a>
<a href="./README_CN.md"><img alt="简体中文版自述文件" src="https://img.shields.io/badge/简体中文-d9d9d9" height="20"></a>
<a href="#self-hosted-deployment"><img alt="Deploy now!" src="https://img.shields.io/badge/Deploy%20now%21-red" height="20"></a>
</div>

# 🔥 Magic - Enterprise-Grade Open-Source AI Agent Platform
## ❓ FAQ

### What is Magic?

Magic is an **enterprise-grade open-source AI Agent platform** built for **security, control, direct business outcomes, and autonomous 24/7 operation**. It combines a Generalist AI Agent + Workflow Engine + IM + Online collaborative office system. Built for OPC (One Person Company) and OPT (One Person Team) operating models.

<div align="center">
<p align="center">
<a href="https://www.magicrew.ai" target="_blank">
<img alt="Static Badge" src="https://img.shields.io/badge/Official Website-301AD2">
</a>
<a href="https://github.com/dtyq/magic/releases">
<img src="https://poser.pugx.org/dtyq/magic/v/stable" alt="Stable Version">
</a>
<a href="https://github.com/dtyq/magic/graphs/commit-activity" target="_blank">
<img alt="Commits last month" src="https://img.shields.io/github/commit-activity/m/dtyq/magic?labelColor=%20%2332b583&color=%20%2312b76a">
</a>
<a href="https://github.com/dtyq/magic/" target="_blank">
<img alt="Issues closed" src="https://img.shields.io/github/issues-search?query=repo%3Adtyq%2Fmagic%20is%3Aclosed&label=issues%20closed&labelColor=%20%237d89b0&color=%20%235d6b98">
</a>
<a href="https://github.com/dtyq/magic/discussions/" target="_blank">
<img alt="Discussion posts" src="https://img.shields.io/github/discussions/dtyq/magic?labelColor=%20%239b8afb&color=%20%237a5af8">
</a>
<a href="https://www.magicrew.ai" target="_blank">
<img alt="Static Badge" src="https://img.shields.io/badge/Built with Magic 🔮-301AD2">
</a>
</p>
</div>
### How does Magic differ from personal AI assistants?

| Feature | Magic | OpenClaw | ChatGPT |
|---------|-------|----------|---------|
| Target | Enterprise | Personal | Individual |
| Data Ownership | Organization | Personal account | Third-party |
| Budget Control | Per-dept/user caps | None | None |
| Sandbox Isolation | ✅ Proprietary | ❌ No | ❌ No |
| Approval Workflow | ✅ Human-in-loop | ❌ No | ❌ No |
| Deliverables | PPT/Dashboards/Excel | Text | Text |
| Expert Agents | ✅ Domain specialists | Skills | Plugins |
| Team Collaboration | ✅ Multi-user projects | ❌ No | ❌ No |
| IM Integration | WeCom/DingTalk/Lark | 25+ channels | None |

<br/>
### 7 Enterprise-Grade Core Capabilities

![MagiCrew](https://public-cdn.magicrew.ai/static/img/3.0/magicrew-publish-head.png)
| Capability | Description |
|------------|-------------|
| **Knowledge Consolidation** | Encapsulate ERP/CRM/databases into digital employees |
| **Results as Deliverables** | PPTs, dashboards, reports, Excel, infinite canvases |
| **Security & Compliance** | Sandbox containers, private VPC, tenant isolation |
| **Human-in-the-Loop** | High-risk operations require approval |
| **Granular Cost Control** | Daily budgets per department/user/Agent |
| **Team Collaboration** | Multi-user projects, real-time visibility |
| **Open Ecosystem** | Compatible with Anthropic Skills + OpenClaw Skills |

[🦞 OpenClaw](https://github.com/openclaw/openclaw) is a great personal AI assistant — connecting all major IMs as conversation channels, supporting any LLM, running autonomously 24/7.
### Personal AI Assistant vs Expert Agents

But when we bring it into an enterprise context, new challenges naturally emerge: data scattered across individual accounts, no budget guardrails, output that stops at plain text, high-risk actions without an approval gate.
**Personal AI Assistant:**
- Every employee gets a dedicated assistant (24/7)
- Connects to calendars, email, internal systems, data, tools
- Mobilizes the right resources and expertise

Magic is built for exactly these challenges: an enterprise AI Agent platform built for **security, control, direct business outcomes, and autonomous 24/7 operation**.
**Expert Agents:**
- Domain specialists (legal, finance, sales, operations)
- Deep and comprehensive within their domain
- Reusable organizational assets

---
### What deployment options are available?

## Stop Tinkering. Start Building Your Enterprise AI Engine.
| Option | Description |
|--------|-------------|
| **Self-Hosted** | Docker deployment via `curl -fsSL https://getmagicrew.sh | bash` |
| **Cloud (China)** | [Magic](https://www.letsmagic.cn) |
| **Cloud (International)** | [MagiCrew](https://www.magicrew.ai) |
| **Enterprise Edition** | Private deployment, dedicated models, custom integration |

Personal AI tools consistently hit the same walls when deployed at scale. Here's how Magic addresses each one:
### Self-Hosted Requirements

- Data locked in employee accounts, gone when they leave → Unified data hub; institutional knowledge stays with the org
- Unpredictable API costs, budget overruns at month-end → Per-department, per-user, per-task budget caps
- Staff using third-party tools, core data at risk → In-house sandbox isolation; data never leaves the trusted boundary
- AI automation deleting files, sending wrong emails → High-risk actions require human approval; humans stay in control
- Can't connect to ERP, CRM → Wrap internal systems as digital employees, accessible to everyone
- Output is just text, still requires manual formatting into decks → Deliver finished artifacts: dashboards, reports, PPTs, Excel files
- **Docker** + curl
- **Platforms**: macOS, Linux (Windows coming soon)
- **Web UI**: http://localhost:38080
- **Teardown**: `magicrew teardown`

## Built for Any Scale
### What deliverables can Magic produce?

Magic isn't only for large enterprises. **From a one-person shop to a 10,000-person organization, it solves the same problem: the output of 100 people at the cost of 1.**
| Deliverable | Description |
|-------------|-------------|
| **PPTs** | Professional presentations |
| **Data Dashboards** | Live operational dashboards |
| **Meeting Summaries** | Structured with owners and deadlines |
| **Reports** | Research reports, financial analysis |
| **Excel Files** | Data tables, calculations |
| **Infinite Canvases** | Image creation, poster design |

It's a natural fit for **OPC (One Person Company)** and **OPT (One Person Team)** operating models — lean assets, zero headcount overhead, fast delivery of tangible results, plug in new capabilities on demand. Whether you're a solo founder or a small team, you can command an entire AI workforce.
### How does sandbox isolation work?

**Solo founders / one-person teams**
- Agents run in **proprietary sandbox containers**
- Isolated from main system in **separate VPC**
- Connected via **private endpoints**
- **Sidecar network proxy** manages traffic per user
- **Complete resource and data isolation** across tenants
- **Plugin security review** catches malicious code

Let AI cover the roles you haven't hired yet — marketing, operations, legal, data analysis, customer support, design, copywriting, finance. Zero labor cost, 24/7 output. Ship dashboards, reports, and draft contracts fast. Need a new capability? Just add an Agent.
### How does Human-in-the-Loop work?

**Enterprises / mid-size teams**
- High-risk operations trigger **approval workflow**
- Routine actions run autonomously
- Destructive actions (deleting data, sending emails) require **explicit human confirmation**
- Decision authority stays with people

Centralized platform to manage AI across the entire org, eliminating data leakage risk. Department budgets are transparent and attributable. Internal expertise is packaged into digital employees, reusable across departments — knowledge never walks out the door. High-risk actions require approval. Sandbox isolation ensures data never leaks.
### What is the cost control mechanism?

---
- **Daily budgets** per department
- **Daily budgets** per user
- **Daily budgets** per Agent
- AI spending becomes **predictable and controllable**
- Every dollar justified

## 7 Enterprise-Grade Core Capabilities
### How does team collaboration work?

### 1. Enterprise Knowledge Consolidation
- Multiple people share a **single project**
- Each owns different modules
- **Real-time visibility** across team
- Expert users can jump into colleague's project
- Progress auto-reported to **WeCom, DingTalk, Lark**

### What Skills ecosystems are supported?

- **Anthropic Skills** ecosystem compatibility
- **OpenClaw Skills** ecosystem compatibility
- Existing tools and skills plug right in
- Zero migration cost

### What languages are supported?

README available in:
- [English](README.md)
- [简体中文](README_CN.md)

### Use Case Scenarios

1. **8 people doing the work of 80** - Cross-border e-commerce automation
2. **Operational truth, any time** - Live dashboards in 30 seconds
3. **Global business, never offline** - 24/7 customer service Expert Agents
4. **Risk intercepted before it happens** - Legal Expert Agent for contracts
5. **Organizational memory compounding** - AI workforce carries collective intelligence
6. **New hire, Day one, Senior-level output** - Ramp time drops to one week
7. **3 people, 10 markets** - Market expansion Expert Agent
8. **Retiring expert's knowledge preserved** - After-sales Expert Agent
9. **Meetings that work** - Auto-compile data, transcribe, structured summaries

Deeply encapsulate fragmented internal systems (ERP / CRM / databases) and domain expertise into **digital employees accessible to everyone**. Break down silos and turn scattered API calls into reusable core digital assets at scale.
### License

![digital-employee-market](https://public-cdn.magicrew.ai/static/img/3.0/digital-employee-market.png)
[Magic Open Source License](LICENSE) - based on Apache 2.0 with additional restrictions.

### 2. Results as Deliverables
### Help Resources

No more stopping at "Chat." The built-in rendering framework transforms AI output directly into finished artifacts — **PPTs, data dashboards, meeting summaries, professional reports, Excel files, infinite canvases (image creation)** — ready to use, no post-processing required.

![openclaw-enterprise-research-report](https://public-cdn.magicrew.ai/static/img/3.0/openclaw-enterprise-research-report.png)
![solution-ppt-demo](https://public-cdn.magicrew.ai/static/img/3.0/solution-ppt-demo.png)
![earnings-call-analysis](https://public-cdn.magicrew.ai/static/img/3.0/earnings-call-analysis.png)
![canvas-poster-design](https://public-cdn.magicrew.ai/static/img/3.0/canvas-poster-design.png)

### 3. Enterprise Security & Compliance

Every Agent runs inside a **proprietary sandbox container**, isolated from the main system in a separate VPC and connected via private endpoints — zero risk of unauthorized access. A Sidecar network proxy manages traffic independently per user, with **complete resource and data isolation across tenants**. A strict **plugin security review** catches malicious code before publishing, keeping enterprise data within a trusted boundary at all times.

### 4. Human-in-the-Loop Control

When an Agent attempts a high-risk operation, an approval workflow is triggered. Routine actions run autonomously; destructive ones — like deleting data or sending emails — **require explicit human confirmation**. Decision authority stays with people.

### 5. Granular Cost Control

An enterprise-grade cost compass lets you set precise **daily budgets per department, per user, and per Agent**. AI spending becomes predictable and controllable — every dollar justified.

### 6. Team-Wide Collaboration

Multiple people share a single project, each owning different modules, advancing in parallel with real-time visibility. Expert users can jump into a colleague's project and help on the spot. **Experience accumulates naturally through projects and compounds across the team.** Progress can be automatically reported to WeCom, DingTalk, or Lark groups — true zero-friction collaboration.

### 7. Open Ecosystem Compatibility

Fully compatible with the **[Anthropic Skills](https://docs.anthropic.com/en/docs/agents-and-tools/tool-use/overview) ecosystem** and the **[OpenClaw Skills](https://github.com/openclaw/openclaw) ecosystem**. Existing tools and skills plug right in — zero migration cost to enterprise-grade.

![skill-creator](https://public-cdn.magicrew.ai/static/img/3.0/skill-creator.png)

---

## Personal AI Assistant + Expert Agents: Everyone Commands an AI Army

Magic brings two complementary capabilities to the enterprise:

**Every employee gets a personal AI assistant**

Think of it as assigning each employee a dedicated 🦞, on call 24/7. The personal assistant doesn't just connect to IM — it **connects to everything**: calendars, email, internal systems, data, tools, and specialist Agents. One instruction is all it takes to mobilize the right resources and expertise — the true meaning of "100-person output at 1-person cost."

**Expert Agents as domain specialists**

Codify the know-how and workflows of legal, finance, sales, operations, and every other function into digital employees. Each Expert Agent is **deep and comprehensive within its domain**, ready to be invoked by users through their personal assistant. A new hire calling in an expert gets senior-level judgment from day one. Domain expertise stops being person-dependent and becomes a reusable organizational asset.

**If you're a decision-maker, picture these scenarios:**

### 8 people doing the work of 80

> A cross-border e-commerce company runs with a team of 8. Each person has a personal AI assistant connected to Expert Agents for product selection, listing, advertising, customer service, logistics, and translation — end-to-end, fully automated. Their competitor does the same work with 80 people. And slower.

### Not monthly reports — operational truth, any time

> 2 AM, a question surfaces: "Can our East China gross margin hold this quarter?" The personal assistant connects to ERP, finance systems, and CRM, calls the finance Expert Agent, and delivers a live dashboard in 30 seconds. Management cadence is no longer held hostage by reporting cycles.

### Global business, never offline

> A New York client sends an urgent email at 3 AM. The customer service Expert Agent is already on it. It knows the product, understands return policy, and communicates fluently in natural English. By morning, what you see isn't a to-do — it's a resolved ticket summary. Time zones are no longer a bottleneck.

### Risk intercepted before it happens

> Every outbound contract passes through the legal Expert Agent before it leaves the building. Risky clauses are flagged and revision-suggested before the client ever sees them, with human approval triggered when needed. Compliance isn't a speed bump — it's a safety net running silently in the background.

### The compounding value of organizational memory

> Every business decision, every problem solved, every client interaction makes the Expert Agents smarter. A year in, your AI workforce carries not one person's experience but the collective intelligence of hundreds — and it never takes leave, never resigns, never withholds what it knows.

### New hire. Day one. Senior-level output.

> It used to take six months to develop a capable project manager. Now, on the first day, the personal assistant connects the new hire to the project management Expert Agent, the industry knowledge base, and the historical case library. Every pitfall has already been documented; every template is already packaged. Ramp time drops from six months to one week, with full productivity from day one.

### 3 people, 10 markets

> A three-person go-global team needs to enter 10 countries. The personal assistant invokes the market expansion Expert Agent to run the full playbook: research local regulations, generate compliant product descriptions and marketing copy, manage listings on local platforms, and track orders and after-sales across regions. Not a few translated paragraphs — a fully operational market entry, end to end.

### The retiring expert whose knowledge never leaves

> The after-sales team has a veteran engineer who can diagnose any failure the moment a customer describes it — what to check, how to fix it, where to source the part. That expertise is now fully encoded in the after-sales Expert Agent: symptom → diagnostic path → solution → parts procurement, a complete decision chain. A junior tech hits a tough problem, asks via their personal assistant, and gets a diagnosis at the veteran's level.

### No more two-hour meetings that go nowhere

> Before: the personal assistant auto-compiles relevant data and open items. During: the meeting Expert Agent transcribes in real time, flags contested points, tracks every action item. Five minutes after it ends: a structured summary — with owners and deadlines — is pushed to every attendee. Meetings become what they should be: decisions, not time sinks.

---

## 🚀 Quick Start

### Self-Hosted Deployment

**Requirements:** Docker + curl. Supports macOS and Linux (Windows support is coming soon).

```bash
curl -fsSL https://getmagicrew.sh | bash
```

The script handles everything — cluster creation, infrastructure, and service deployment. Once complete:

- Web UI: **http://localhost:38080**
- Teardown: `magicrew teardown`

→ [Full deployment guide](./docs/en/development/deploy/docker.md)

### Cloud Service

Prefer not to self-host? Use the cloud version — sign up and go, zero configuration:

- **China**: [Magic](https://www.letsmagic.cn)
- **International**: [MagiCrew](https://www.magicrew.ai)

### Enterprise Edition

We offer enhanced management capabilities and features for teams and enterprises, including private deployment, dedicated model integration, and deep custom integration with your internal systems. [Email us](mailto:[email protected]?subject=[GitHub]Business%20License%20Inquiry) to discuss your needs.

---

## Contribution

To contribute code, see the [Contribution Guide](https://github.com/dtyq/magic/blob/master/CONTRIBUTING.md) / [贡献指南（中文）](https://github.com/dtyq/magic/blob/master/CONTRIBUTING_CN.md). You're also welcome to support Magic through social media, events, and conferences — the project grows with community involvement.

## Security Vulnerabilities

If you discover a security vulnerability, email [[email protected]](mailto:[email protected]). All security issues are handled promptly.

## 📄 License

This repository is licensed under the [Magic Open Source License](LICENSE), based on Apache 2.0 with additional restrictions.

## 🙏 Acknowledgements

Thanks to all developers who have contributed to Magic!

[![Star History Chart](https://api.star-history.com/svg?repos=dtyq/magic&type=Date)](https://star-history.com/#dtyq/magic&Date)
- **Issues**: [GitHub Issues](https://github.com/dtyq/magic/issues)
- **Discussions**: [GitHub Discussions](https://github.com/dtyq/magic/discussions)
- **Email**: [email protected] (security), [email protected] (business)