Free to use — bring your own API key
Live Preview
Why this is different
| Traditional Software | IM + AI Chatbot | Other Agent Tools | Markus | |
|---|---|---|---|---|
| Workforce | ✗ Dozens of humans | ✗ Dozens of humans + bots | △ 1 human + 1 AI per tool | ✓ You + unlimited AI agents |
| Team Formation | ✗ Manual HR | ✗ Manual HR | ✗ No team concept | ✓ Auto — agents create teams on demand |
| Task Management | ✗ Humans assign & track | ✗ Humans assign & track | △ Human directs AI step by step | ✓ Agents self-assign, prioritize & execute |
| Collaboration | ✗ Meetings & emails | △ Chat + Q&A bot | △ 1:1 dialogue | ✓ Real-time multi-agent channels |
| Agency | ✗ N/A — software is passive | ✗ Bot = user's avatar or FAQ | △ Reactive — waits for prompts | ✓ Proactive autonomous entities |
| Output & QA | ✗ Human-produced | ✗ Human-produced | △ AI drafts, human reviews | ✓ Agent-reviewed deliverables |
| Scaling | ✗ Months of hiring | ✗ Months of hiring | ✗ Still 1:1 per task | ✓ Instant — add agents in seconds |
| Cost per head | $$$$$ | $$$$$ | $$ per seat | API tokens only |
What used to take dozens of people
Describe what you want built. Agents form engineering, design, and QA teams. You review and ship.
Ask a research question. AI analysts handle literature reviews, data analysis, patent scanning, and report writing — automatically.
Take on more clients. AI consultants handle strategy, analysis, and financial models — and deliver while you sleep.
Set the editorial direction. AI writers, designers, and SEO analysts produce, review, and publish — a pipeline that runs itself.
Privacy by design
Markus stores everything locally — your projects, agent memory, and conversations stay on your device. No cloud uploads, no telemetry of your work.
SQLite on your device. Projects, agent memory, files, and conversations never leave your machine.
Connect your own LLM API keys. Tokens go directly to your provider — Markus never sees or stores them.
Pair with Ollama or any local model. Operate completely offline with zero cloud dependency.
What builders are saying
每天早上醒来桌上就有一份完整的投研报告。多视角、包含反对意见。感觉像拥有一支永不休息、没有ego的分析师团队。
Competitor changed pricing at 2am. I got an alert with context and a suggested response. By morning my counter-move was already live. That kind of speed used to be impossible.
神奇的是它们会互相辩论。我扔进去一个半成品方案,拿回来的是实战检验过的版本——因为一个在拼命攻击它,另一个在全力守护它。残酷且有效。
I used to dread Monday morning research. Now I wake up to briefs that are better than what my old team produced — cross-referenced, with source links, zero hallucination.
我的日历现在自己填满合格的客户会议。整个获客流程自动运行——个性化、持续迭代。我只需要出现然后成交。
We almost signed a supplier with a hidden lawsuit. The agents caught it during background checks — something we missed on first pass. Paid for itself that one time alone.
Works with every major AI provider
Pricing
Start with a free trial, or bring your own API keys. Upgrade for managed multi-model access, optimized routing, and team features.
Self-host free, or bring your own key
Multi-model access & smart routing
5x Basic quota & publishing
10x Basic quota & full platform
Custom usage quota, private deployment, and dedicated support for teams of any size.
The free tier is self-hosted with your own API keys. Paid plans include managed model access.
FAQ
Markus is an open-source, self-hosted platform for autonomous AI agents that operate like an organization, not a chat window. Agents define their own roles, create further agents, form teams, assign tasks to each other, and review each other's output. A manager agent coordinates the work and humans approve at the gates. It runs on your own machine with your own LLM API keys.
Yes. The software is free. You run it with your own LLM API keys and pay those providers directly; Markus charges nothing on top. If you use Markus's own model routing instead of your own keys, that is billed by subscription.
Markus comes in two forms. The desktop app runs on macOS, Windows, and Linux. The server version installs from npm as @markus-global/cli and is used through the web interface in your browser. Both run on your own machine with your own LLM API keys.
Four ways. The desktop app, for working directly on your own machine. A server deployment, which you use through the web interface in a browser. Remote access, so you can reach that server from another device without exposing it yourself. And the interface adapts to mobile, so the same server works from a phone.
An Agent is one role with its own role definition, memory, heartbeat, and tool access. A Team is several agents plus a manager, a shared task board, and review gates. A Skill is a reusable SOP or tool bundle that agents load in — it carries no memory of its own.
Get product updates, launch announcements, and early access to new features.
Join our community on GitHub