Z.AI, GLM, and ZCode

Open Interpreter includes providers for the global Z.AI platform, its GLM Coding Plan, and the corresponding Zhipu AI services in China. It also includes a native zcode harness for GLM coding workflows.

Choose the Right Provider

Z.AI documents separate general and Coding Plan endpoints. Use the Coding Plan provider only for an eligible subscription; using the general endpoint does not consume Coding Plan quota. Confirm that your client and use case comply with Z.AI's current Coding Plan usage policy before relying on subscription quota.

Start With the Built-In Provider

Export the key for your account, start Open Interpreter, then use /model to select the matching provider and a GLM model:

export ZAI_API_KEY="..."
interpreter

For a direct Z.AI Coding Plan launch:

ZAI_API_KEY="..." interpreter \
  -c 'model_provider="zai-coding-plan"' \
  -m glm-5.2

For the general Z.AI API, change the provider to zai. Users of the China service should use zhipuai or zhipuai-coding-plan with ZHIPU_API_KEY.

The bundled providers use Z.AI's OpenAI-compatible Chat endpoint. With no explicit harness, they use Open Interpreter's generic Chat agent surface.

Use the ZCode Harness

zcode reproduces the ZCode-shaped system prompt, Messages request format, tools, todos, plan controls, skills, session context, and subagent behavior in Open Interpreter's native Rust runtime. It requires an Anthropic Messages-compatible provider; selecting zcode on the built-in Chat provider does not activate that Messages route.

Z.AI officially provides an Anthropic-compatible Coding Plan endpoint at https://api.z.ai/api/anthropic. To use it, add a Messages provider to ~/.openinterpreter/config.toml:

model_provider = "zai-zcode"
model = "glm-5.2"
harness = "zcode"

[model_providers.zai-zcode]
name = "Z.AI ZCode"
base_url = "https://api.z.ai/api/anthropic"
env_key = "ZAI_API_KEY"
wire_api = "messages"
env_http_headers = { Authorization = "ZAI_AUTHORIZATION" }

Then expose the bearer header without writing the secret into the config file:

export ZAI_API_KEY="..."
export ZAI_AUTHORIZATION="Bearer $ZAI_API_KEY"
interpreter

Open Interpreter sends the ZCode Messages request to /v1/messages. The additional authorization environment variable matches Z.AI's documented bearer-token authentication while ZAI_API_KEY remains the provider's required secret source.

See Z.AI's official Coding Plan quickstart and tool endpoint guide for current endpoint and plan-eligibility details.

Choose a GLM Model

Use /model rather than maintaining a private list. The picker is generated from maintained provider sources and includes current GLM model IDs for the selected service. For example, the bundled Z.AI Coding Plan catalog currently includes glm-5.2, glm-5.1, glm-5-turbo, and lower-cost models.

Z.AI may update server-side model mappings and plan eligibility independently of Open Interpreter. Check its model-switching guide when a model is unavailable or consumes a different quota multiplier.

Chat or ZCode?

Do not set wire_api = "messages" on the OpenAI-compatible /paas/v4 endpoint, and do not point the Chat provider at /api/anthropic. The endpoint, wire API, and harness must agree.

Editors and SDKs

Both configurations can be used through interpreter acp in ACP-compatible editors and through Open Interpreter's Codex SDK compatibility. The selected provider and harness remain part of the Open Interpreter configuration.

Troubleshooting

• A 401 or 403 usually means the wrong key, endpoint, region, or authorization header is in use.
• If Coding Plan quota is not being used, confirm that the provider is zai-coding-plan or that the ZCode setup points to /api/anthropic.
• If /harness shows zcode but behavior looks generic, confirm the active provider has wire_api = "messages".
• If a model is absent, open /model and choose one offered for that exact provider instead of copying a model ID from another Z.AI region or plan.