Plugin Manifest

1. Introduction

The OpenClaw Plugin system allows hardware devices to integrate with the OpenClaw AI platform as channel plugins. Each plugin declares the channels it provides and the configuration it requires via a manifest file.

Nexting is the reference implementation of a hardware channel plugin — a wearable input device that connects via BLE through an iPhone companion app to the OpenClaw Gateway and Agent remote-control path.

Version 0.2.0, July 2026. Plugins run as part of the OpenClaw Gateway process.

2. Manifest Format

Every plugin MUST include an openclaw.plugin.json file at its package root. This manifest declares the plugin identity, the channels it provides, and the JSON Schema for its configuration.

{
  "id": "pinclaw",
  "channels": ["pinclaw"],
  "name": "Nexting",
  "description": "Hardware voice interface channel for OpenClaw",
  "channelConfigs": {
    "pinclaw": {
      "schema": {

3. Configuration Schema

The plugin reads its configuration from the channels.pinclaw section of openclaw.json. The TypeScript interface defines the available fields:

interface NextingAccountConfig {
  enabled?: boolean;     // default: true
  wsPort?: number;       // default: 18790
  authToken?: string;    // required for operation
  relay?: {              // Nexting Cloud relay connection
    enabled: boolean;
    token: string;
    url?: string;        // default: wss://api.nexting.ai
  };
}

Single Account Configuration

The simplest configuration uses a single default account. All fields are placed directly under channels.pinclaw:

{
  "channels": {
    "pinclaw": {
      "enabled": true,
      "wsPort": 18790,
      "authToken": "your-secret-token"
    }
  }
}

Multi-Account Configuration

To connect multiple devices, use the accounts object. Each key is a unique account ID, and each value is a NextingAccountConfig:

{
  "channels": {
    "pinclaw": {
      "accounts": {
        "device-1": { "wsPort": 18790, "authToken": "token-1" },
        "device-2": { "wsPort": 18791, "authToken": "token-2" }
      }
    }
  }
}

4. Adapters

The plugin exposes five adapter interfaces that the Gateway uses to manage the plugin lifecycle, route messages, and query status.

Config Adapter

Responsible for discovering accounts, resolving their full configuration, and reporting their readiness.

Outbound Adapter

Handles agent-to-device message delivery. The Gateway calls sendText when the AI produces a response that needs to reach a hardware device.

The sendText method routes to NextingWsServer.sendToDevice(). If the server instance is unavailable (e.g., during startup), the message is dropped (an error is logged) and sendText returns without throwing.

{ channel: "pinclaw", messageId: string }

Status Adapter

Tracks per-account runtime state. The Gateway queries this to display plugin health in dashboards.

{
  accountId: string,
  enabled: boolean,
  configured: boolean,
  running: boolean,
  lastError: string | null
}

Security Adapter

Controls which entities can send messages through the plugin channel.

Gateway Adapter

The startAccount method is the main lifecycle entry point. It runs as a long-lived process within the Gateway:

5. Outbound Delivery

The full message routing flow from AI agent response to device delivery:

6. Tool Definitions

The plugin ships three server tools, auto-discovered from plugin/src/tools/ and injected into TOOLS.md as <server_tool> prompt tools.

generate_image

Generate an image from a text prompt using AI.

{
  "name": "generate_image",
  "description": "Generate an image from a text prompt using AI. Returns the image URL that can be displayed to the user via markdown.",
  "parameters": {
    "prompt": { "type": "string", "required": true, "description": "Detailed English description of the image to generate" },
    "size": { "type": "string", "required": false, "description": "\"1024x1024\" (default), \"1792x1024\" (landscape), or \"1024x1792\" (portrait)" }
  }
}

generate_audio

Generate speech audio from text using OpenAI TTS.

{
  "name": "generate_audio",
  "description": "Generate speech audio from text using OpenAI TTS. Returns a URL to the audio file.",
  "parameters": {
    "text": { "type": "string", "required": true, "description": "The text to convert to speech" },
    "voice": { "type": "string", "required": false, "description": "\"alloy\" (default), \"echo\", \"fable\", \"onyx\", \"nova\", or \"shimmer\"" },
    "speed": { "type": "number", "required": false, "description": "Speed from 0.25 to 4.0 (default 1.0)" }
  }
}

memory_edit

Read, write, or list memory files.

{
  "name": "memory_edit",
  "description": "Read, write, or list memory files (~/clawd/memory/*.md).",
  "parameters": {
    "action": { "type": "string", "required": true, "description": "\"list\", \"read\", \"write\", or \"append\"" },
    "filename": { "type": "string", "required": false, "description": "File name, e.g. \"2026-03-11.md\". Required for read/write/append." },
    "content": { "type": "string", "required": false, "description": "Content to write or append. Required for write/append." }
  }
}

7. Inbound Events

The plugin receives inbound events from the client over its WebSocket connection. The main inbound message types are:

8. Permission Model

9. Cron Integration

Cron jobs deliver scheduled messages to devices through the Gateway's cron system. The plugin server exposes REST endpoints whose cron-proxy forwards a cron.add RPC (announce: true, with an optional [DELIVERY:NOTIFY|SILENT|RESULT] tag prefix) to the gateway.

Cron Endpoints

Create Job Request

{
  "message": "Time to take a break!",
  "at": "30m",
  "channel": "pinclaw",
  "deleteAfterRun": false
}

When channel is "pinclaw", the cron-proxy forwards a cron.add RPC to the gateway with announce: true (optionally prefixed with a [DELIVERY:NOTIFY|SILENT|RESULT] tag). The job fires in the gateway, not in the plugin.

Delivery Flow

10. Lifecycle Hooks

startAccount(ctx)

The primary lifecycle hook. The Gateway calls this method for each enabled and configured account. The context object provides:

Startup Flow

1. Read gateway port from ctx.cfg.gateway.port  (default: 18789)
2. Read gateway token from ctx.cfg.gateway.auth.token
3. Create NextingWsServer instance
4. Call server.start()
5. Connect Gateway WebSocket RPC (Ed25519 auth)
6. Set status: running = true
7. Await abort signal (blocks — keeps gateway alive)
8. On abort: server.stop(), running = false

Gateway Auto-Reconnect

If the Gateway WebSocket disconnects, the plugin automatically reconnects after a 3-second delay. This ensures the RPC channel recovers from transient network issues without manual intervention.