Integrate GLM Models in Openclaw

Table of contents: [Show]

Quick Start

GLM models from Z.AI bring powerful reasoning capabilities to your Openclaw setup through a streamlined integration process.

Access to state-of-the-art Chinese LLMs with competitive performance benchmarks.
Configuration complexity often blocks developers from leveraging these models effectively.
A working GLM integration that auto-configures thinking modes and tool streaming.

Generate an API key from the Z.AI console. This key authenticates your requests to the GLM model family.

Run the interactive onboarding wizard for the fastest setup:

openclaw onboard --auth-choice zai-api-key

For automated deployments, use non-interactive mode:

openclaw onboard --zai-api-key "$ZAI_API_KEY"

Edit your ~/.openclaw/openclaw.json for persistent settings:

{
  "env": { "ZAI_API_KEY": "sk-..." },
  "agents": { 
    "defaults": { 
      "model": { "primary": "zai/glm-5" } 
    } 
  }
}

After configuration, test your setup:

openclaw ask "Explain quantum computing in simple terms"

Expected response: A clear, structured explanation generated by the GLM model.

GLM-4.x models enable thinking mode by default. Disable with:

--thinking off

Or configure permanently in openclaw.json:

agents.defaults.models["zai/<model>"].params.thinking = false

Tool call streaming is enabled by default. To disable:

params.tool_stream = false

Alias normalization: Openclaw accepts z.ai/* and z-ai/*, auto-converting to zai/*.
Coding endpoints: Use --auth-choice zai-coding-global or zai-coding-cn for specialized coding plans.
Cerebras alternative: Set primary model to cerebras/zai-glm-4.7 for Cerebras-hosted GLM.

GLM integration expands your model options with competitive alternatives to Western LLMs, complete with automatic feature configuration.