Client Integrations

GLM 5 exposes documented OpenAI-compatible Chat Completions and Responses APIs. A client must send requests to the GLM 5 base URL, use a GLM 5 key, and select a model ID that GET /models returns.

Verify the API first

Before configuring an IDE or agent client, verify the key and endpoint directly:

curl https://glm5.app/api/v1/models \
  -H "Authorization: Bearer $GLM5_API_KEY"

If this request fails, fix the key, paid API balance, or endpoint first. A client-specific setup cannot repair an API request that fails here.

OpenCode

OpenCode supports custom OpenAI-compatible providers. Add your credential with /connect and choose Other, then configure GLM 5 in your OpenCode config. Current OpenCode releases may use either the provider/npm/options schema or the newer providers/package/settings schema, so follow the schema generated by your installed version.

A current OpenAI-compatible provider configuration looks like this:

{
  "$schema": "https://opencode.ai/config.json",
  "providers": {
    "glm5": {
      "name": "GLM5",
      "env": ["GLM5_API_KEY"],
      "package": "@opencode-ai/ai/providers/openai-compatible",
      "settings": {
        "baseURL": "https://glm5.app/api/v1"
      },
      "models": {
        "glm-5.3-flash": { "name": "GLM 5.3 Flash" },
        "glm-5.3": { "name": "GLM 5.3" }
      }
    }
  }
}

After saving, select the provider/model in OpenCode. Keep the API key in the local credential store or GLM5_API_KEY, never in source control.

Oh My Pi (omp)

omp (Oh My Pi) supports custom OpenAI-compatible providers through ~/.omp/agent/models.yml. Use the Responses protocol for GLM5:

providers:
  glm5:
    baseUrl: https://glm5.app/api/v1
    api: openai-responses
    apiKey: GLM5_API_KEY
    models:
      - id: glm-5.3-flash
        name: GLM 5.3 Flash
        contextWindow: 1000000
        maxTokens: 8192

Set GLM5_API_KEY in the same shell that starts omp. To make the model the default for new sessions, add this to ~/.omp/agent/config.yml:

modelRoles:
  default: glm5/glm-5.3-flash

Verify discovery and start a session:

omp models glm5
omp

On Windows native, the equivalent files are %USERPROFILE%\\.omp\\agent\\models.yml and %USERPROFILE%\\.omp\\agent\\config.yml. In WSL, use the Linux paths unless you explicitly configure a shared OMP agent directory. Keep apiKey as the environment-variable name and never commit the real key.

Codex Desktop / CLI

Codex custom providers use the Responses API. GLM5 supports the Responses endpoint and translates the request into the same metered GLM5 model runtime. Configure the provider at the user level so it applies to both Codex CLI and the Codex desktop app:

model = "glm-5.3-flash"
model_provider = "glm5"

[model_providers.glm5]
name = "GLM5"
base_url = "https://glm5.app/api/v1"
env_key = "GLM5_API_KEY"
wire_api = "responses"

Set the key in your local shell before starting Codex:

export GLM5_API_KEY="sk-glm5-..."
codex --model glm-5.3-flash

On macOS and Linux, export affects the current terminal session. On Windows, use PowerShell or Command Prompt instead:

$env:GLM5_API_KEY = "sk-glm5-..."
codex --profile glm5

set GLM5_API_KEY=sk-glm5-...
codex --profile glm5

The profile file has the same name on all platforms; only the home directory changes:

The Responses adapter supports text streaming, function calls, and the standard input/output item shapes. Keep the full input history in the request; GLM5 does not persist Codex response state server-side.

Verify the setup

Inside Codex, run /status and confirm that the active model is glm-5.3-flash. If the request fails, check the key without printing it:

test -n "${GLM5_API_KEY:-}" && echo "GLM5_API_KEY is set"

An MCP startup incomplete warning, such as a Linear OAuth warning, is a separate MCP connection issue and does not indicate a GLM5 API failure.

Return to the original Codex provider

The GLM5 setup is a profile overlay. To return to your normal Codex provider, stop the current session and start Codex without the profile:

codex

This keeps your original ~/.codex/config.toml, MCP servers, and OpenAI login unchanged. You do not need to delete the base Codex config.

To disable only the GLM5 profile while keeping a recoverable backup:

mv ~/.codex/glm5.config.toml ~/.codex/glm5.config.toml.disabled
unset GLM5_API_KEY

If GLM5_API_KEY was added to ~/.zshrc or ~/.bashrc, remove that one export line and reload the matching shell file. Do not remove ~/.codex/config.toml; it may contain your existing provider, MCP, permissions, and notification settings.

Browser, MCP, and API errors are different

These messages do not mean the GLM5 API configuration is wrong:

Cursor

Cursor is configured in its settings interface, not through a project config file. In Cursor Settings > Models:

1. Enable Use OpenAI API Key and paste your sk-glm5-... key.
2. Enable Override OpenAI Base URL (when using key).
3. Set the Base URL to https://glm5.app/api/v1.
4. Add or select a model ID returned by GET /models, such as glm-5.3 or glm-5.3-flash.
5. Select that model for a supported chat request.

If the installed Cursor build or account does not offer the OpenAI Base URL override, it cannot directly use this endpoint. Use OpenCode, Continue, or an SDK instead.

Claude Code

Claude Code uses Anthropic's Messages protocol rather than the OpenAI Chat Completions protocol exposed by GLM 5, so do not point Claude Code directly at https://glm5.app/api/v1.

To use GLM 5 from Claude Code, place an Anthropic-compatible translating gateway between Claude Code and GLM 5. The gateway must accept Anthropic Messages requests, translate them to OpenAI-compatible Chat Completions, and send them to https://glm5.app/api/v1 with your GLM 5 key. Keep both the gateway and key local/private.

If you do not need Claude Code specifically, OpenCode is the simpler direct path for the GLM 5 OpenAI-compatible API.

Continue for VS Code

Continue is a VS Code extension with an explicit OpenAI-compatible provider configuration. Add this to its config.yaml, replacing the key placeholder:

name: GLM5 Config
version: 0.0.1
schema: v1

models:
  - name: GLM 5.3 Flash
    provider: openai
    model: glm-5.3-flash
    apiBase: https://glm5.app/api/v1
    apiKey: <YOUR_GLM5_API_KEY>
    roles:
      - chat
      - edit
      - apply

Use a model ID returned by GET /models. After saving, begin with one short chat request before using the client for edits or longer agent workflows.

OpenAI SDK

Use the standard OpenAI client with the GLM 5 base URL:

import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.GLM5_API_KEY,
  baseURL: 'https://glm5.app/api/v1',
});

const response = await client.chat.completions.create({
  model: 'glm-5.3-flash',
  messages: [{ role: 'user', content: 'Review this function for edge cases.' }],
  max_completion_tokens: 800,
});

console.log(response.choices[0]?.message?.content);

Common setup failures

For the supported request fields, see Chat Completions. Do not rely on client-specific fields that are not documented there.