Migrating from OpenAI

Change the URL, key, and model while checking API compatibility.

Model protocol

In the OpenAI SDK, replace base_url and the key, then set an exact Guard model. Check other fields against the selected family and API contract: compatibility does not imply every OpenAI feature is supported.

diff

  from openai import OpenAI
  client = OpenAI(
-     api_key=OPENAI_KEY,
+     api_key=GUARD_KEY,
+     base_url="https://api.guardrelay.ai/v1",
  )

Set a Claude model such as claude-sonnet-5-5 instead of an OpenAI model, and check parameters against Chat Completions.

Kimi, Composer and Grok

For another family, set its exact model and check the fields supported by the selected API. The address and key stay shared.

Family	What goes in `model`	Endpoint
Claude	`claude-sonnet-5-5`, `claude-opus-5`, `claude-haiku-5-5`	`/v1/chat/completions`, `/v1/messages`, `/v1/responses`
Kimi	`k3`, `k3-256k`, `kimi-for-coding`, `kimi-for-coding-highspeed`	`/v1/chat/completions`, `/v1/messages`, `/v1/responses`
Composer	`composer-2.5`, `composer-2.5-fast`	`/v1/chat/completions`, `/v1/messages`, `/v1/responses`
Grok	`cursor-grok-4.7-high` is the recommended default; the other seven Grok 4.7 ids and eight Grok 4.6 ids are in the catalog	`/v1/chat/completions`, `/v1/messages`, `/v1/responses`

Move an OpenAI client to GLM

For Chat Completions or Responses, change the Base URL and model ID. Keep other fields only when the matching GLM endpoint page documents them.

Field	New value
Base URL	`https://api.guardrelay.ai/glm/v1`
API key	`GUARD_API_KEY` (`gd-…`)
Model	`glm-5.3` or `glm-5.3-flash`
Output budget	`max_tokens: 4096` or `max_output_tokens: 4096`

javascript

const client = new OpenAI({
  apiKey: process.env.GUARD_API_KEY,
  baseURL: "https://api.guardrelay.ai/glm/v1",
  maxRetries: 0,
});
const response = await client.responses.create({
  model: "glm-5.3",
  input: "Reply with OK",
  max_output_tokens: 4096,
  store: false,
});

Responses is stored by default; send store: false when continuation through previous_response_id is not needed.
background: true is unsupported.
Built-in search uses the exact GLM objects on Web search, not an arbitrary schema from another API.
Files/Conversations/Assistants/Batch and WebSocket are outside the published GLM contract.

The response model field preserves the Z.AI value; billing and Usage API remain attached to the exact public ID requested by the client.

Migrating from OpenAI

Kimi, Composer and Grok

Move an OpenAI client to GLM

Sections

On this page