Best for: Teams and enterprises that want shared, governed AI agents that research, analyze, and produce ready-to-use reports and presentations
MagiCrew is an open-source enterprise AI agent platform where teams deploy specialized digital workers for research, analysis, and deliverable-ready reporting.
MagiCrew is sold as an AI agent platform built around capabilities rather than conversation. You brief agents in plain language and they work toward a concrete deliverable, whether that is a research report, a live dashboard, a presentation, or an Excel export. The idea is to remove the need for employees to juggle several separate AI subscriptions.
The product pairs a generalist personal assistant with expert agents covering finance, legal, customer support, sales, data analytics, and project management. Those roles are packaged as reusable digital workers a whole team can call on, and the surrounding workspace adds workflow automation, messaging, and a collaborative office.
Open-source core on GitHub at dtyq/magic
Generalist assistant plus prebuilt expert agents
Digital workers aimed at real business outputs
Cloud and self-hosted deployment options
How MagiCrew Works
Behind the scenes, an orchestrator agent takes a request, breaks it into steps, and dispatches specialist agents in parallel. Each agent works within the tools, skills, models, and data you have connected, and the results are compiled through a rendering framework into a finished file rather than a plain-text reply.
High-impact or destructive operations pause for human approval instead of running unattended. Admins can also cap spending per department, per user, or per agent, so the cost of a heavy workflow is visible before it arrives on a bill.
Orchestrator splits tasks and dispatches specialist agents in parallel
Deliverables rendered directly into slides, dashboards, reports, and spreadsheets
Human approval gate for high-risk actions
Per-department, per-user, and per-agent budget limits
MagiCrew Features
The platform centers its pitch on seven enterprise capabilities. The ones that matter most to buyers are outlined below.
AI Digital Workers
A digital worker is an agent with a defined role. You can wrap internal systems such as ERP, CRM, or databases into a reusable digital worker, or package documented expertise so the whole organization can query it. Ready-made expert agents cover finance, legal, support, sales, analytics, and project management, and a worker marketplace makes them available to every employee.
Multi-Agent Collaboration
An orchestrator coordinates specialist agents on a single objective, runs the pieces in parallel, and merges the output. The practical effect is that research, analysis, and deck-building no longer happen sequentially in one chat thread, which is the usual bottleneck with a single assistant.
Research and Analysis
Official examples lean heavily on research and analysis: deep web research that aggregates sources into a cited report, earnings-call analysis, live data dashboards, and meeting notes. Separate playbooks cover deep research, data analysis, and meeting summaries for teams that want a starting template.
Deliverable-Ready Outputs
Output does not stop at text. A built-in rendering framework converts agent results into presentations, professional reports, data dashboards, Excel files, meeting summaries, and image canvases, so teams get a file they can ship rather than a draft that needs reformatting.
Skills and Integrations
Skills are reusable capabilities that extend what an agent can do. The platform is compatible with the Anthropic Skills and OpenClaw Skills ecosystems, so existing skills import without migration work, and a Skill Creator lets you define new ones through conversation. ERP and CRM systems can be connected and wrapped as digital workers, and API or MCP endpoints are supported.
Anthropic Skills and OpenClaw Skills compatibility
Conversational Skill Creator
ERP/CRM wrapping and MCP/API integrations
Plugin security review before skills are published
Team and Enterprise Controls
On the management side, the platform offers three-tier budget controls, cost attribution down to department, user, and agent, role-based sharing, and project collaboration where multiple people advance modules in parallel. Progress can be reported automatically into WeCom, DingTalk, or Lark groups.
Agents run in sandbox containers isolated from the main system in a separate VPC, and every user's traffic is managed through its own sidecar network proxy.
MagiCrew Pricing
Pricing is structured as a monthly plan plus point-based usage. Each paid plan includes a monthly point allowance, and extra agent workloads draw points billed at about $0.34-0.35 per 100. The prices shown on the official site are promotional, marked as roughly 66-67% off a reference price, and annual billing is offered with its own discount.
There is a free tier on sign-up with one parallel task, limited features, select models, and 1GB of cloud storage. Unused monthly points do not carry over, and point top-ups are reserved for paid plans. Check the live pricing page for current numbers, since promotional rates can change.
Compare Plans & Pricing
Free tier on sign-up. Paid plans range from $9.99 to $99.99 per month at current promotional rates, plus point-based usage at about $0.34-0.35 per 100 points.
Free
$0/ month
Free on sign-up. Select models, limited features, and 1GB cloud storage.
Solo founders and one-person teams use it to cover marketing, operations, legal, and data-analysis roles without adding headcount. Enterprises get a central platform where digital workers are shared, budgeted, and reviewed rather than living in personal chat tabs.
If you only need a single chatbot or a light writing assistant, the platform is more than you need. It pays off when the goal is a shared workforce that produces finished work and can be governed.
Security and Privacy
On security, the vendor describes each agent running in a proprietary sandbox container, isolated from the main system in a separate VPC and connected through private endpoints, with tenant isolation across workspaces. Human approval is required for high-risk actions such as deleting data or sending email, and a plugin security review is applied before skills are listed.
On privacy, the service is operated by Lighthouse Engine under a privacy policy last updated March 26, 2025. It states that personal information collected within mainland China is stored on servers located in China and is not currently transferred across borders. The public site makes no GDPR, SOC 2, or ISO 27001 certification claims, so organizations with data-residency obligations should review the policy and terms before adopting the cloud edition.
Proprietary sandbox isolation, VPC separation, and per-user sidecar proxies
Human approval for high-risk operations
Plugin security review before skills are published
Privacy policy by Lighthouse Engine (March 2025), China-based storage, no certification claims
Open Source and Self-Hosting
The source is published on GitHub at dtyq/magic under the Magic Open Source License, which the repository describes as Apache 2.0 based with additional restrictions. That is an important detail: it is a source-available-style custom license rather than plain Apache 2.0, so review it if your policy requires a standard permissive license. The repository carries roughly 5,000 stars and 500 forks.
Self-hosting runs through a Docker-based installer that provisions a single-node or cluster deployment on macOS or Linux, with Windows support listed as coming soon. The cloud has separate editions for China and the international market, and a paid enterprise edition adds private deployment and dedicated model integration.
GitHub: dtyq/magic (~5,000 stars)
Magic Open Source License: Apache 2.0 based with restrictions
Self-host via Docker on macOS/Linux, Windows coming later
Separate China and international cloud editions, plus enterprise private deployment
Alternatives to MagiCrew
A few lighter options sit alongside it in this directory. OpenClaw is a self-hosted, open-source agent gateway that puts a personal assistant inside messaging apps such as WhatsApp, Telegram, Discord, Slack, and iMessage, and it is notably relevant here because the platform is built to reuse the OpenClaw Skills ecosystem. AutoGPT is an open-source autonomous agent that plans multi-step goals, while AgentGPT runs goal-based agents straight from the browser.
The platform is the heavier commitment of the group: it adds digital workers, deliverable outputs, budget controls, and office tooling on top of raw agents. Pick a lighter tool if you just want one assistant running on your chat apps, and return here when you need a governed, shared workforce.
OpenClaw
AI AgentsFreemium
Best for: Developers and power users who want a private, self-hosted AI assistant they can message from any chat app
An open-source, self-hosted AI agent gateway that brings a personal assistant into WhatsApp, Telegram, Discord, Slack, and iMessage, with persistent memory, tools, and skills.
AutoGPT is an open-source autonomous AI agent that plans multi-step goals, runs them with tools and web access, and reports back without constant human prompting.
Best for: Free browser-based AI agents for task automation
AgentGPT runs autonomous AI agents straight from your browser — free to start, no API key required, and it breaks any goal into sub-tasks it executes step by step.
The platform is an unusual combination of open source and enterprise controls, and that is both its main strength and its main catch. You get shared agents with approval workflows, granular budgets, and finished deliverables, but you also inherit a custom license, point-based usage costs, promotional pricing that can change, and a China-based cloud and privacy setup.
Teams that can self-host and control their own data get the most out of it. Teams that only want a quick chatbot will find the setup and cost model heavier than they need.
Pros
Open-source core under the Magic Open Source License, based on Apache 2.0
Digital workers that hand back finished deliverables, not just chat replies
Multi-agent collaboration with an orchestrator and specialist agents
Human approval required for high-risk actions
Granular budget controls and cost attribution by department, user, and agent
Compatible with Anthropic and OpenClaw Skills, plus a conversational Skill Creator
Cons
Custom license is Apache-2.0 based with additional restrictions, not plain Apache 2.0
Point-based usage adds variable cost on top of subscriptions, and unused monthly points do not roll over
Displayed prices are promotional, roughly 66-67% off the listed reference rate
Self-hosting currently supports macOS and Linux via Docker, with Windows coming later
Cloud privacy setup is China-based with no GDPR, SOC 2, or ISO 27001 certification claim
Best For
Recommended use cases and scenarios where MagiCrew shines.
An open-source AI agent platform that deploys digital workers for research, analysis, and reporting, producing deliverables such as reports, presentations, dashboards, and Excel files rather than plain chat replies.
How does MagiCrew work?
You describe a goal in plain language, an orchestrator agent breaks it into steps and dispatches specialist agents in parallel, and results are rendered into finished files. High-risk actions wait for human approval.
Is MagiCrew open source?
Yes. The core is published on GitHub under the Magic Open Source License, described as Apache 2.0 based with additional restrictions.
Can MagiCrew be self-hosted?
Yes. A Docker-based installer provisions a self-hosted instance on macOS or Linux, with Windows support listed as coming soon.
How much does MagiCrew cost?
A free tier is available on sign-up. Paid plans run from $9.99 to $99.99 per month at current promotional rates, and heavy usage adds points billed at about $0.34-0.35 per 100.
What can MagiCrew agents do?
Depending on configured tools, skills, models, and permissions, agents can research the web, analyze data, draft contracts, summarize meetings, and build reports, dashboards, presentations, and Excel deliverables.
Is MagiCrew secure?
The vendor outlines sandboxed agents with VPC isolation, tenant isolation, and plugin security review, and requires human approval for high-risk operations. The public policy makes no GDPR, SOC 2, or ISO 27001 claims.
Reviews & Ratings
Loading reviews...
Explore More AI Agents Tools
Browse the full AI Tools Vault directory to compare MagiCrew with every ai agents tool, or line up your shortlist side by side.
AutoGPT is an open-source autonomous AI agent that plans multi-step goals, runs them with tools and web access, and reports back without constant human prompting.
