openclaw-clawglassos

Community OpenClaw channel plugin for ClawGlassOS — the Even Realities G2 smart-glasses companion app in this repo. Not an Even Realities or OpenClaw official package.

What this is

A Channel Plugin in the same slot as the bundled qqbot, googlechat, msteams extensions in OpenClaw. It exposes the glasses as a chat channel (clawglassos) so OpenClaw can:

• receive user voice / text from the glasses, and
• push AI replies / notifications back to the glasses display.

Wiring

┌─────────────┐  WSS   ┌────────────────────────┐   inbound dispatch   ┌──────────┐
│  WebView    │ <────> │ ws-server (plugin)     │ ───────────────────► │ OpenClaw │
│ (on iPhone) │        │  :8787 /ws            │ ◄─────────────────── │  core    │
│  + G2 mic   │        └────────────────────────┘  outbound.sendText   └──────────┘
└─────────────┘

• Channel id: clawglassos
• Default endpoint: ws://<host>:8787/ws?token=<token>&device=<id>
• Inbound frames (client → server):
  • {type:"hello", deviceId, label?} — sent once, right after socket open.
  • {type:"text", text} — typed input.
  • {type:"voice", text, lang?, streamId?} — client-side STT result; treated like text but tagged as voice for OpenClaw rendering.
  • {type:"audio_start", streamId, encoding?, sampleRate?, channels?, lang?}
    • binary PCM frames + {type:"audio_end", streamId?} — server-side STT path. Defaults: pcm_s16le, 16 kHz, mono.
  • {type:"ping"}.
• Outbound frames (server → client):
  • {type:"reply", messageId, text} — AI reply.
  • {type:"transcript", streamId?, text, final} — server STT echo of the user's utterance (only emitted on the audio_start path).
  • {type:"display", text}, {type:"status", text}, {type:"pong"}, {type:"error", text}.

MVP scope

• Open WSS endpoint
• Token gate on connection
• Text round-trip (WebView → OpenClaw → reply → WebView)
• Audio path stub: receives PCM, currently surfaces a placeholder ([voice message]) for STT. Real STT slot reserved in src/stt.ts.

Status

Plumbing is in place; the actual injection of a user message into OpenClaw's session store follows the same pattern as the bundled qqbot extension (runtime.gateway + buildInboundContext + dispatchOutbound). The dispatch site is marked TODO(plugin-runtime) in src/inbound.ts and will be wired against the real PluginRuntime next.

Build

npm install
npm run build

Then load it into your OpenClaw install per your plugin loader conventions (externally-installed plugins use defineChannelPluginEntry; the bundled defineBundledChannelEntry path is only for in-repo extensions).