OpenCode with Kimi K3, GLM 5.3 and GLM 5.3 Flash: Add an EU Custom Provider

OpenCode works with any OpenAI-compatible API through a custom provider in opencode.json. Point it at the Nodion AI Gateway (https://api.nodion.ai/v1) and you can code with Kimi K3, GLM 5.3 and GLM 5.3 Flash. The gateway routes only to providers based in the EU, all serving with zero data retention. You need no account with those providers: OpenCode calls everything with your Nodion API key.

A version note first: OpenCode now ships in two lines. OpenCode 1 is the current GitHub release (v1.19.5, October 2026), documented at opencode.ai/docs. OpenCode 2 is a separate install with its own docs and different native config keys. The example below uses the V1 config format. It works in the current release, and OpenCode 2 accepts it too; its migration guide says "The V1 config format remains supported."

What you need

• A Nodion account and a project with AI Inference.
• An API key: an inference key from AI Inference → Keys (it starts with nodion_ai_ and is shown once), or a key you create on your AI Coding Plan. Keys for the Nodion platform API do not work here.
• Credit in the prepaid wallet, unless the key is on a coding plan.

Export the key in the shell that starts OpenCode:

export NODION_AI_KEY=nodion_ai_your_key

Add the provider to opencode.json

Use the global file ~/.config/opencode/opencode.json or an opencode.json in the project root:

{
  "$schema": "https://opencode.ai/config.json",
  "model": "nodion/moonshot/kimi-k3",
  "provider": {
    "nodion": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Nodion",
      "options": {
        "baseURL": "https://api.nodion.ai/v1",
        "apiKey": "{env:NODION_AI_KEY}"
      },
      "models": {
        "moonshot/kimi-k3": {
          "name": "Kimi K3",
          "limit": { "context": 1048576, "output": 32768 }
        },
        "z-ai/glm-5.3": {
          "name": "GLM 5.3",
          "limit": { "context": 1048576, "output": 131072 }
        },
        "z-ai/glm-5.3-flash": {
          "name": "GLM 5.3 Flash",
          "limit": { "context": 1048576, "output": 131072 }
        }
      }
    }
  }
}

• npm: OpenCode's docs name @ai-sdk/openai-compatible for OpenAI-compatible providers on /v1/chat/completions. Stay on this route. The docs point to @ai-sdk/openai for /v1/responses, but that AI SDK provider sets store to true by default, and Nodion's Responses endpoint stores nothing.
• apiKey: {env:NODION_AI_KEY} reads the key from the environment. If the variable is not set, OpenCode puts in an empty string, so you get an auth error rather than a config error.
• models: the keys are the Nodion model ids, slash included. OpenCode refers to a model as provider id plus model id, so Kimi K3 becomes nodion/moonshot/kimi-k3. Its docs show the same nesting with lmstudio/google/gemma-3n-e4b.
• limit: for standard providers OpenCode pulls context and output limits from models.dev. Nodion has no entry there (checked 29 October 2026), so set both by hand from the catalog. They tell OpenCode how much context is left.

Instead of {env:...} you can store the key with /connect: scroll to Other, enter the provider id nodion and paste the key. You still need the provider block in opencode.json.

Sources, checked October 2026: OpenCode providers, models, config, OpenCode 2 providers, AI SDK OpenAI provider, models.dev.

Select the model

The top-level model is the default. In the TUI, /models lists your custom provider and its models. On the command line you switch with -m, opencode run works without the TUI, and opencode models nodion shows which models OpenCode knows for the provider:

opencode -m nodion/z-ai/glm-5.3
opencode run -m nodion/z-ai/glm-5.3-flash "Explain the build setup of this repo"
opencode models nodion

Sources: OpenCode CLI, models.

The three models side by side

Values from the public Nodion catalog on 29 October 2026:

Model	Id	Context	Max output	Input	Reasoning efforts
Kimi K3	moonshot/kimi-k3	1,048,576	32,768	Text, image, video	low, high, max
GLM 5.3	z-ai/glm-5.3	1,048,576	131,072	Text	low, high, max
GLM 5.3 Flash	z-ai/glm-5.3-flash	1,048,576	131,072	Text, image, video, file	low, high, max

All three support tools, JSON mode, structured outputs and reasoning. GLM 5.3 takes text only, so screenshots and other images need Kimi K3 or GLM 5.3 Flash. The reasoning efforts are the values the API accepts for reasoning_effort. Prices and providers are on each model's page.

Use a coding plan key

AI Coding Plans (Go, Pro and Max) pay for chat models and tools. OpenCode's requests are chat completions, so they count against the plan. The project owner buys one plan per developer and assigns it to a project member, who creates keys on that plan in AI Inference. Put that key into NODION_AI_KEY; the config stays the same.

Usage counts against a 5-hour, a weekly and a monthly window. When one is full, requests continue from the wallet only if the plan allows wallet use. Keys without a plan pay from the prepaid wallet. Details on AI Coding Plans.

Troubleshooting

• 401: NODION_AI_KEY is not set in the shell that started OpenCode (the {env:...} value is then empty), or it holds a platform key. Use an inference key (nodion_ai_…) or a plan key.
• 404 model not found: a key under models is not an exact Nodion id. curl https://api.nodion.ai/v1/models lists the ids and needs no key.
• Model missing in /models: check that it is listed under provider.nodion.models. opencode models nodion shows what OpenCode sees.
• OpenCode cannot tell how much context is left: limit.context and limit.output are missing for that model.
• 402: the wallet cannot cover the request, or a spend limit was reached.
• 503: no healthy route for the model matches the project's AI settings, for example because its providers are excluded there.

Sources: AI API reference: errors and limits, AI Inference docs, AI Coding Plans.

Next steps

• Codex instead of OpenCode: Codex CLI with EU models.
• Aider, Cline and Continue: Aider, Cline and Continue with EU models.
• Every field of chat completions on Nodion: AI API reference.
• A monthly package for coding agents: AI Coding Plans.