Build Your AI Workforce. Orchestrate Autonomous Employees. Turn OpenClaw into a Self-Operating Company.

中文 | English

Mimiclaw is the Workforce Layer of OpenClaw.

It is not another agent. It is not a wrapper. It is not a multi-agent demo.

Mimiclaw is the system that allows OpenClaw to:

• Create employees
• Assign responsibilities
• Parallelize execution
• Supervise progress
• Enforce approval boundaries
• Control digital labor at scale

It turns intelligence into organization.

Every intelligent system eventually faces the same limitation:

A single agent cannot scale like a company.

Mimiclaw solves this.

It enables OpenClaw to:

• Spawn independent AI employees
• Assign them structured roles
• Allow them to communicate
• Execute tasks in parallel
• Report progress in stages
• Escalate asset-sensitive operations for approval
• Be destroyed, replaced, or reconfigured at will

This is not agent chaining. This is structured digital labor.

OpenClaw becomes the CEO.

Employees become:

• Researchers
• Coders
• Operators
• Analysts
• Executors

All autonomous. All isolated. All supervised.

                ┌────────────────────┐
                │      OpenClaw      │
                │  (Supervisor Core) │
                └─────────┬──────────┘
                          │
             ┌────────────┼────────────┐
             │            │            │
        ┌────▼────┐  ┌────▼────┐  ┌────▼────┐
        │Employee 1│  │Employee 2│  │Employee 3│
        │(picoclaw)│  │(picoclaw)│  │(picoclaw)│
        └──────────┘  └──────────┘  └──────────┘

OpenClaw allocates vision. Mimiclaw builds execution.

Mimiclaw is built around five principles:

Each employee is:

• An independent picoclaw process

• Running in its own workspace

• Equipped with its own:

  • Skills
  • Tools
  • API keys
  • Wallet abstraction
  • Memory

• Fully destroyable

Employees cannot spawn other employees.

Hierarchy remains controlled.

Employees execute concurrently.

• No preemption.

• Priority equals sorting.

• User defines maximum concurrent employees.

• When limit is reached:

  • OpenClaw may destroy or replace agents.

This models resource allocation inside a company.

All communication happens locally via WebSocket.

• OpenClaw ↔ Employees
• Employee ↔ Employee

Messages are JSON-based.

Employees:

• Receive structured tasks
• Send progress checkpoints
• Report final results
• Request asset approvals

OpenClaw logs everything.

This creates a transparent communication graph.

Employees may propose asset actions.

They may never sign.

Workflow:

1. Employee proposes asset operation
2. OpenClaw asks user
3. User approves
4. OpenClaw signs
5. Result returned to employee

All signatures are centralized.

Authority is never delegated.

OpenClaw is:

• Task allocator
• Signature authority
• Health monitor
• Lifecycle controller
• WebSocket hub

Employees are workers.

OpenClaw is control.

An employee is considered alive only if:

1. Process exists
2. WebSocket heartbeat is active
3. Agent self-report is valid

Failure in any dimension triggers supervision logic.

{
  "task_id": "uuid",
  "from": "openclaw",
  "to": "employee-1",
  "priority": 5,
  "goal": "Analyze smart contract risk",
  "inputs": {},
  "tools_allowed": ["bash", "curl"],
  "assets_flag": false,
  "approval_required": false,
  "timestamp": 1700000000
}

{
  "task_id": "uuid",
  "status": "in_progress",
  "progress": 0.45,
  "message": "Parsing contract ABI...",
  "timestamp": 1700000100
}

{
  "task_id": "uuid",
  "type": "asset_request",
  "action": "transfer",
  "payload": {
    "to": "0x...",
    "amount": "1.5 ETH"
  }
}

{
  "task_id": "uuid",
  "status": "completed",
  "result": {},
  "timestamp": 1700000200
}

Mimiclaw includes a workforce dashboard.

It visualizes:

• Employee topology
• Task allocation
• Priority levels
• Progress checkpoints
• Communication logs
• Supervisor approval events

Topology Example:

[OpenClaw]
    │
    ├── [Researcher-1]
    ├── [Coder-1]
    └── [Operator-1]

Full message tracing is preserved.

This is not only execution.

It is observability.

OpenClaw assigns models per employee or per task.

• Model switching is centralized.
• Employees do not choose models.
• Strategy remains in supervisor.

This prevents chaos.

Mimiclaw exposes full workforce control:

• create employee
• delete employee
• assign task
• broadcast task
• list employees
• get status
• fetch logs

The entire company is programmable.

• Autonomous AI companies
• DAO operations
• Parallel research teams
• Code generation pipelines
• Automated trading assistants
• Continuous security auditing

• Not a token system
• Not a governance layer
• Not a resource quota manager
• Not recursive agent spawning
• Not cost-optimized cloud infrastructure

It is an execution architecture.

Mimiclaw represents a shift:

From: Single intelligent system

To: Structured autonomous workforce

From: Tool calling

To: Organized digital labor

From: AI assistant

To: AI company

Design complete. Protocol defined. WebSocket architecture in progress. Process supervision documentation pending.

MIT