How do I connect ZCode to LMU AI?

Open **Settings → Model Providers**, click "Add provider" at the bottom of the provider list on the left, give it any name (e.g. `LMU AI`), set the Anthropic endpoint to `https://api.lmuai.com` and the OpenAI endpoint to `https://api.lmuai.com/v1`, enter the `sk-` key generated in the LMU AI console, then save, click "Add model" to enter a model ID manually (e.g. `claude-opus-5`), and switch the enable toggle on.

Do the endpoints need a `/v1` suffix?

The two protocols differ. The **Anthropic endpoint** takes `https://api.lmuai.com` (**no** `/v1` — the client appends `/v1/messages` itself), while the **OpenAI endpoint** takes `https://api.lmuai.com/v1` (**with** `/v1`). This matches the official examples for third-party providers, where Anthropic-style entries use the bare host and OpenAI-style entries go down to `/v1`.

LMU AI models don't appear in the model picker

Custom providers don't fetch a model list automatically. After saving the provider, click "**Add model**" and enter the model ID by hand (e.g. `claude-opus-5`, `glm-5.2`), then switch the enable toggle on. It will then be selectable in the model picker in the chat box.

Can I change the context window for a custom provider's models?

Yes. In **Settings → Model Providers**, open a model and expand "**Advanced**" to set the context window and max output tokens for that individual model. LMU AI's mainline Claude models support a 1M context window — use a model ID with the `[1M]` suffix (e.g. `claude-opus-5[1M]`) to enable it.

Why are the thinking-effort levels different from the official endpoint?

This is documented ZCode behaviour — a third-party deployment of a same-named model may accept different effort levels than the official endpoint, and custom models typically offer only an on/off pair or no levels at all, depending on the model configuration. Functionality is unaffected; just use it as normal.

Does ZCode's HTTP proxy setting affect requests to LMU AI?

Yes. The HTTP proxy in ZCode's settings isn't a global proxy, but model requests are among the traffic that goes through it. If a proxy is stopping you from connecting, turn it off and retry — and note that the `HTTP_PROXY` / `HTTPS_PROXY` environment variables apply too, and ZCode needs a restart after changing them.

ZCode

Add a custom provider in the ZCode editor to connect to the LMU AI API — one entry holds both the Anthropic and OpenAI endpoints, giving you Claude, GPT and Chinese models.

ZCode is an Agentic Development Environment (ADE) desktop app built for long-horizon tasks. It supports custom providers for any model service that speaks the Anthropic or OpenAI protocol — this page covers connecting it to LMU AI.

Install ZCode

Download and install the build for your operating system from the official site:

Website: https://zcode.z.ai

Add LMU AI as a provider

Open Settings → Model Providers and click "Add provider" at the bottom of the provider list on the left, then fill in:

Field	Value
Name	anything you like, e.g. `LMU AI`
Anthropic endpoint	`https://api.lmuai.com`
OpenAI endpoint	`https://api.lmuai.com/v1`
API Key	the `sk-` key generated in the LMU AI console

The two endpoints take different /v1 forms

Anthropic endpoint: https://api.lmuai.com — no /v1; the client appends /v1/messages itself
OpenAI endpoint: https://api.lmuai.com/v1 — must include /v1

Mixing them up produces a 401 or 404. Both addresses can live in the same provider entry — ZCode uses whichever protocol matches the model you pick.

Add models

Custom providers don't fetch a model list automatically, so add them by hand:

Save the provider configuration
Click "Add model"
Enter the model ID manually, e.g. claude-opus-5
Switch the enable toggle on

Repeat for each model you want. Common model IDs:

Model ID	Notes
`claude-opus-5`	Claude Opus 5 (flagship, recommended)
`claude-sonnet-5`	Claude Sonnet 5 (balanced)
`claude-haiku-4-5`	Claude Haiku 4.5 (fast)
`gpt-5.6-sol`	GPT-5.6 Sol
`glm-5.2`	GLM-5.2
`qwen3.8-max-preview`	Qwen 3.8 Max Preview
`deepseek-v4-pro`	DeepSeek V4 Pro
`kimi-k3`	Kimi K3

Not sure which model to enter? Go to the Model Gallery to see every available model ID and copy it with one click.

Enable the 1M context window

Mainline models such as claude-opus-5 and claude-sonnet-5 support a 1M token context window — add the [1M] suffix to the model ID to enable it:

claude-opus-5[1M]

Enter the full ID with the suffix using the "Add model" steps above.

The context window can be tuned per model

In Settings → Model Providers, open a model and expand "Advanced" to set the context window and max output tokens for that individual model. Both are editable for a custom provider's models.

Leaving max output tokens empty is recommended (it then follows the model's own limit). Note that output space shares the same context as your conversation history, so the higher you set it, the less history can be carried and the sooner automatic compaction kicks in.

Verify the setup

Restart ZCode once configured, select your newly added model in the chat box's model picker, and send a simple instruction to test. A normal reply means you're set.

Can't connect? Check the proxy first

The HTTP proxy in ZCode's settings isn't a global proxy, but model requests do go through it. If a proxy is breaking the connection, turn it off and retry. The HTTP_PROXY / HTTPS_PROXY environment variables apply as well, and ZCode needs a restart after you change them.

Notes

LMU AI is a third-party / custom provider, so create it with "Add provider" at the bottom of the list — don't modify the built-in vendor entries
The Anthropic endpoint takes https://api.lmuai.com (no /v1); the OpenAI endpoint takes https://api.lmuai.com/v1 (with /v1)
For the API Key, enter the sk- key generated in the LMU AI console
Custom providers require adding each model by hand and switching its enable toggle on before it shows up in the picker
A third-party deployment of a same-named model may offer different thinking-effort levels than the official endpoint; custom models typically expose only an on/off pair, or none at all
If you hit image-related errors with a custom model, confirm the model itself accepts image input and declare its image capability in the model configuration

Last updated: 2026-09-09