How to use glm model

[PangoSea (kapaseker)]

I have registered an account for the GLM model, and there are several free models available in it. I want to know how to use Koog to connect to the GLM model instead of using the predefined model types provided by Koog.

[Faraz Ahmed (faraz152)]

GLM models (from Zhipu AI) are not built into Koog directly, but you can connect to them easily through the OpenRouter integration, which Koog already supports out of the box. OpenRouter acts as a unified API gateway and carries all major GLM variants (e.g. zhipuai/glm-4-flash, zhipuai/glm-4, zhipuai/glm-4-plus).

Steps

1. Add the OpenRouter client dependency to your build.gradle.kts:

implementation("ai.koog:prompt-executor-openrouter-client:<version>")

2. Get an OpenRouter API key at openrouter.ai (free tier available). You do not need a separate Zhipu AI account — OpenRouter handles routing to Zhipu on your behalf.

3. Define the model and create an executor:

import ai.koog.prompt.executor.clients.openrouter.OpenRouterLLMClient
import ai.koog.prompt.llm.LLMCapability
import ai.koog.prompt.llm.LLMProvider
import ai.koog.prompt.llm.LLModel

// Define GLM-4 Flash using its OpenRouter model ID
val glm4Flash = LLModel(
    provider = LLMProvider.OpenRouter,
    id = "zhipuai/glm-4-flash",   // free-tier, fast
    capabilities = listOf(
        LLMCapability.Temperature,
        LLMCapability.Tools,
        LLMCapability.Completion,
    )
)

val client = OpenRouterLLMClient(apiKey = System.getenv("OPEN_ROUTER_API_KEY"))

4. Use it in an agent:

val agent = AIAgent(
    promptExecutor = simpleAIExecutor(client),
    llmModel = glm4Flash,
    systemPrompt = "You are a helpful assistant.",
    toolRegistry = ToolRegistry.EMPTY,
)

val response = agent.run("Hello from Koog + GLM!")
println(response)

Available GLM model IDs on OpenRouter

Model	OpenRouter ID	Notes
GLM-4 Flash	zhipuai/glm-4-flash	Free tier, fast
GLM-4	zhipuai/glm-4	Flagship
GLM-4 Plus	zhipuai/glm-4-plus	Enhanced reasoning
GLM-4 Long	zhipuai/glm-4-long	1M context window

Browse all Zhipu models at openrouter.ai/models?q=zhipuai.

Why OpenRouter instead of a direct Zhipu client?

Koog's OpenRouterLLMClient is already fully wired for streaming, tool calls, embeddings, and structured output. OpenRouter lets you skip writing a custom LLMClient subclass entirely and use any GLM model with a single API key.