GLM Realtime

Zhipu GLM Realtime voice provider plugin for OpenClaw

Install

openclaw plugins install clawhub:@shushushv/openclaw-glm-realtime

OpenClaw GLM Realtime

A Zhipu GLM Realtime voice provider plugin for OpenClaw. It adds a streaming, server-VAD voice-chat provider (zhipu) so you can talk to your OpenClaw agent through Zhipu's GLM Realtime API.

Requirements

OpenClaw gateway >= 2026.6.9
A Zhipu BigModel API key with GLM Realtime access (glm-realtime-plus tier)

Install

bash

openclaw plugins install clawhub:@shushushv/openclaw-glm-realtime

Configure

Set your API key via environment variable:

bash

export ZHIPU_API_KEY="your-zhipu-api-key"

Or configure it in openclaw.json under the provider config. Recognized keys:

Key	Aliases	Default	Notes
`apiKey`	`api_key`, `ZHIPU_API_KEY`	—	Falls back to the `ZHIPU_API_KEY` env var
`model`	—	`glm-realtime-plus`	`glm-realtime-plus` recommended; `glm-realtime` is an alias to a paid backend
`voice`	`speaker`	`tongtong`	GLM voice name

The provider registers as zhipu (label: Zhipu GLM Realtime), with aliases glm, glm-realtime, zhipu-realtime.

How it works

Connects to wss://open.bigmodel.cn/api/paas/v4/realtime with Bearer auth.
Streams microphone audio as 24 kHz mono PCM frames, each tagged with a client_timestamp so the server-side VAD can detect end-of-speech.
Uses turn_detection: server_vad — the server auto-commits and responds; no manual commit/response is needed.
Supports tool calls via response.function_call_arguments.done.

Known limitations

No proactive greeting: in server-VAD mode GLM ignores a response.create that has no preceding user audio, so an opening line cannot be triggered.
"Speaking" UI indicator: a general gateway-relay limitation, not specific to this provider.

License

MIT