Overview
Markus Platform Overview
An open-source AI workforce platform — your AI agents work as a real team, collaborate, review each other, and ship verifiable results. 中文 · Back to Docs Center
Markus in One Sentence
A single AI assistant is like a smart intern who forgets everything and says “done” too early. Markus exists to fix exactly this:
Markus gives your AI an organization — memory, peers, review, and 24/7 momentum — so work actually ships instead of stopping at “looks done.”
You describe a goal in plain language; Markus assembles a role-based team (researcher, developer, reviewer, writer, ops), breaks the work into tasks, runs specialists in parallel, and makes every delivery pass through peer review before it reaches you. Your team keeps working while you sleep.
What You Can Run on Markus
| Area | Example |
|---|---|
| Research & analysis | Scan competitors, compile a competitive analysis, draft a go-to-market strategy |
| Software engineering | Build a feature end-to-end — requirements, code, tests — with built-in peer review |
| Content & publishing | Draft, edit, review, and schedule articles, reports, and social posts |
| Operations | Daily briefings, issue triage, scheduled monitoring, blocker escalation |
| Data & reporting | Pull data, analyze it, and deliver finished reports on a schedule |
| Personal research | Deep-dive any topic and get a structured, cited deliverable back |
The same team keeps learning: the longer it runs, the more it knows your context, the smarter it gets.
Why a Team Beats a Copilot
| Single AI agent | Markus AI team | |
|---|---|---|
| Scale | One task at a time | Parallel work across specialist roles |
| Memory | Context evaporates when the session ends | Three-layer persistent memory, auto-consolidated between sessions |
| Initiative | Waits for your prompt, every time | Heartbeat patrols tasks 24/7 — works while you sleep |
| Quality | ”Done” is self-reported | Peers review, catch mistakes, and gate every delivery |
| Visibility | 10 agents = 10 windows | One dashboard, one audit trail |
| Accountability | No guardrails | Progressive trust levels, submit–review–merge, emergency pause |
You manage a workforce, not individual prompts.
Core Capabilities
🧠 Three-layer persistent memory
Procedural (how to do things), semantic (what it knows), episodic (what happened). Knowledge accumulates across sessions and consolidates automatically — your team gets measurably smarter the longer it runs.
⏰ Heartbeat-driven initiative
Agents don’t wait to be prompted. The heartbeat scheduler patrols open tasks, processes async completions, and surfaces blockers — work keeps shipping after you close the laptop.
🛡️ Progressive trust levels
Probation → standard → trusted → senior. Agents earn scope and authority as they demonstrate reliability.
✅ Quality gates
A formal submit → review → merge lifecycle. Nothing reaches you unreviewed; peers catch what self-reporting misses.
📜 Full audit trail
Every action is logged: who, what, when, and why. Reproduce results or investigate incidents without guessing.
🔌 An open ecosystem, not a walled garden
- 80,000+ community skills, plug and play —
markus skill importauto-detects and normalizes skills from skills.sh / Claude Code, SkillHub / ClawHub, OpenClaw, SOUL.md, AgentScope, and MCP servers. No rewriting, no lock-in. - Any LLM provider — Anthropic, OpenAI, Google, DeepSeek, MiniMax, Ollama, OpenRouter, SiliconFlow, Moonshot — with automatic failover and per-task routing.
- Any MCP server plugs straight into the agent tool layer.
Open Source & Data Ownership
This one matters most:
- Open source — Apache-2.0, on github.com/markus-global/markus. Fork it, study it, customize it for your needs.
- Local, data yours — runs entirely on your infrastructure. SQLite by default. No mandatory cloud, no data leaving your network unless you choose remote access.
- Bring your own keys — connect any LLM provider; credentials live in your deployment, never in a third-party cloud.
- Isolated workspaces — each agent works in its own sandboxed workspace; no cross-talk between runs.
- Optional remote access — Cloudflare Tunnel, Tailscale, FRP, or ngrok to manage your team from anywhere.
Comparisons
Comparisons are by category, not product-bashing. Markus differentiates on four things: open source + local deployment + data ownership + an ecosystem you’re not locked into.
vs single-agent code/writing assistants (Claude Code, Cursor, Copilot…)
| Dimension | Single assistant | Markus |
|---|---|---|
| Shape | One session, one task | A multi-role team, working in parallel |
| Memory | Forgets when the session ends | Three-layer persistent memory |
| Quality | Self-reported “done” | Peer review, quality gates |
| Momentum | Waits for your prompt | Heartbeat 24/7 |
vs multi-agent orchestration frameworks (CrewAI, AutoGen…)
| Dimension | Framework | Markus |
|---|---|---|
| Runtime | Bring your own runtime & infra | Full agent runtime built in: shell, files, git, browser, GUI, MCP out of the box |
| Governance | You implement it | Requirements → tasks → review → merge, governance built in |
| Delivery | You build the UI/audit | Built-in dashboard, audit trail, deliverables |
vs commercial AI office/agent platforms (WorkBuddy, Coze…)
| Dimension | Commercial platform | Markus |
|---|---|---|
| Open source | Closed, black box | Open source (Apache-2.0), fork & customize |
| Deployment | Cloud-hosted, data on their servers | Local deployment, data ownership stays yours |
| Models | Tied to vendor ecosystem | Any LLM provider, BYOK, no lock-in |
| Community assets | In-platform marketplace only | 80,000+ community skills, interoperable ecosystem |
Where to Go Next
- To get hands-on: Docs Center → Quick Start or Quickstart →
- Deployment details: Installation →, First Agent →, Create Task →
- Role-based learning paths: see the Docs Center →