ClawLoops Layered AI Runtime Architecture

System Design · flowchart diagram · Apache-2.0

A layered architecture diagram for ClawLoops, illustrating the separation of concerns across entry, platform, orchestration, runtime, and AI model access l

Source: https://github.com/machinesociety/ClawLoops/blob/7d2c1214f45e98d01c1e3dc8b8c519a1c2e53488/README/README.md
Curated by machinesociety
AI Platform System Design Runtime Orchestration Layered Architecture Model Access Security Isolation Traefik

Mermaid source

flowchart TB
    A([Browser])

    subgraph L1[Entry Layer]
        direction TB
        B[Traefik]
    end

    subgraph L2[Platform Layer]
        direction TB
        D[ClawLoops Control Plane]
    end

    subgraph L3[Orchestration Layer]
        direction TB
        E[Runtime Orchestrator]
        F[Runtime Manager]
    end

    subgraph L4[Runtime Layer]
        direction TB
        G[Per-user OpenClaw Runtime]
    end

    subgraph L5[Model Layer]
        direction TB
        H[Model Service / Credential Proxy]
        I[LiteLLM]
        J[vLLM / Ollama / Upstream Providers]
    end

    K[(PostgreSQL)]

    A -->|Ingress| B -->|Session check| D
    D -->|Desired state| E -->|Lifecycle orchestration| F -->|Start / Stop / Recover| G
    D -->|Model access| H --> I --> J
    D -->|Platform data| K

    classDef edge fill:#eff6ff,stroke:#2563eb,stroke-width:2px,color:#0f172a;
    classDef plane fill:#dcfce7,stroke:#16a34a,stroke-width:2px,color:#0f172a;
    classDef runtime fill:#f3e8ff,stroke:#7c3aed,stroke-width:2px,color:#0f172a;
    classDef model fill:#ffe4e6,stroke:#db2777,stroke-width:2px,color:#0f172a;
    classDef data fill:#f8fafc,stroke:#64748b,stroke-width:2px,color:#0f172a;

    class A,B edge;
    class D plane;
    class E,F,G runtime;
    class H,I,J model;
    class K data;

What this diagram shows

This diagram illustrates a multi-layered system architecture for an AI runtime platform called ClawLoops. It depicts the flow from browser ingress through Traefik, a central control plane, runtime orchestration, and dedicated per-user runtimes, culminating in access to various AI model services via a proxy and LLM providers. It highlights the distinct responsibilities of each layer for security, scalability, and governance.

When to use it

Use this pattern when designing complex AI platforms requiring strict separation of concerns, secure per-user environments, scalable model access, and robust runtime lifecycle management. It's suitable for MLOps platforms, AI-as-a-Service, or environments where multiple users interact with isolated AI compute.

How to adapt it for your project

This architecture can be adapted by swapping specific components (e.g., different ingress controllers, alternative orchestration tools like Kubernetes, or other LLM providers). The runtime layer can be customized for specific compute needs (GPU, CPU). Additional layers for data storage, monitoring, or logging can be integrated. The model access layer can be extended to support fine-grained access control for different models.

Key concepts