OpenClaw

flowchart TD
  subgraph User["User Layer"]
    CLI["CLI / TUI"]
    WebUI["Web UI"]
    MobileApps["Mobile Apps"]
  end
  subgraph Gateway["Gateway Control Plane"]
    Boot["Boot / Init"]
    Auth["Auth & Rate Limiting"]
    Router["Message Router"]
    Sessions["Session Manager"]
  end
  subgraph Channels["Channel Layer"]
    WA["WhatsApp"]
    TG["Telegram"]
    Slack["Slack"]
    Discord["Discord"]
    More["25+ more..."]
  end
  subgraph Core["Core Engine"]
    Agents["Agent Orchestrator"]
    Commands["Command Processor"]
    AutoReply["Auto-Reply Engine"]
    CtxEngine["Context Engine"]
  end
  subgraph AI["AI / LLM Layer"]
    Providers["Provider Adapters"]
    Tools["Tool Execution"]
    MCP["MCP Bridge"]
    ImgGen["Image Generation"]
  end
  subgraph Plugins["Plugin Ecosystem"]
    PluginSDK["Plugin SDK"]
    Bundled["Bundled Plugins"]
    ExtDir["Extension Directory"]
  end
  subgraph Infra["Infrastructure"]
    Config["Configuration"]
    Store["Data Store"]
    Cron["Cron Scheduler"]
    Secrets["Secrets Vault"]
  end

  CLI --> Gateway
  WebUI --> Gateway
  MobileApps --> Gateway
  Gateway --> Channels
  Channels --> Core
  Core --> AI
  Core --> Plugins
  AI --> Providers
  Plugins --> MCP
  Gateway --> Infra
  Core --> Infra

sequenceDiagram
    participant User as User (WhatsApp/Telegram/...)
    participant Ch as Channel Extension
    participant GW as Gateway Router
    participant Cmd as Command Processor
    participant Agent as Agent Orchestrator
    participant LLM as LLM Provider
    participant Tool as Tool / MCP Server

    User->>Ch: Send message
    Ch->>GW: Normalized message (ChatMeta)
    GW->>GW: Auth check + rate limit
    GW->>Cmd: Route to command processor
    Note over Cmd: Check for slash command prefix
    alt Slash command (e.g. /model)
        Cmd->>Cmd: Execute command handler
        Cmd-->>Ch: Command response
    else Natural language
        Cmd->>Agent: Forward to agent
        Agent->>Agent: Build context (history + system prompt)
        Agent->>LLM: Stream request (OpenAI/Anthropic/...)
        loop Streaming tokens
            LLM-->>Agent: Token chunk
            Agent-->>Ch: Forward chunk to user
        end
        opt Tool call requested
            Agent->>Tool: Execute tool (MCP/function)
            Tool-->>Agent: Tool result
            Agent->>LLM: Continue with tool result
            LLM-->>Agent: Final response
        end
        Agent-->>Ch: Complete response
    end
    Ch-->>User: Deliver response