OpenCode.

OpenCode is an open source terminal agent. You declare NaN as one more provider and from then on you can switch models from inside the agent itself.

You can skip all of this The NaN CLI writes this configuration for you: Setup tab, e to paste your key, space to mark OpenCode, c to apply.

Configuration

Write this into ~/.config/opencode/opencode.json to have it in every project, or into an opencode.json at the root of the project if you only want it there:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "nan": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "NaN",
      "options": {
        "baseURL": "https://api.nan.builders/v1",
        "apiKey": "sk-your-key"
      },
      "models": {
        "deepseek-v4-flash": {
          "name": "DeepSeek V4 Flash",
          "limit": { "context": 1048575, "output": 32768 },
          "modalities": { "input": ["text", "image"], "output": ["text"] }
        },
        "glm5.3-flash": {
          "name": "GLM 5.3 Flash",
          "limit": { "context": 1048576, "output": 32768 },
          "modalities": { "input": ["text", "image"], "output": ["text"] }
        },
        "qwen3.8-flash": {
          "name": "Qwen 3.8 Flash",
          "limit": { "context": 262144, "output": 32768 },
          "modalities": { "input": ["text", "image"], "output": ["text"] }
        },
        "mimo-v2.5": {
          "name": "Xiaomi MiMo V2.5",
          "limit": { "context": 1048576, "output": 32768 },
          "modalities": { "input": ["text", "image", "audio"], "output": ["text"] }
        },
        "gemma4": {
          "name": "Gemma 4",
          "limit": { "context": 262144, "output": 65536 },
          "modalities": { "input": ["text", "image"], "output": ["text"] }
        },
        "qwen3.6": {
          "name": "Qwen 3.6",
          "limit": { "context": 262144, "output": 65536 },
          "modalities": { "input": ["text", "image"], "output": ["text"] }
        },
        "glm5.3": {
          "name": "GLM 5.3 (premium)",
          "limit": { "context": 1048576, "output": 32768 },
          "modalities": { "input": ["text"], "output": ["text"] }
        }
      }
    }
  },
  "compaction": {
    "auto": true,
    "prune": true,
    "reserved": 50000
  }
}

This is the config for the 7 LLM models NaN serves: deepseek-v4-flash, glm5.3-flash, qwen3.8-flash, mimo-v2.5, gemma4, qwen3.6 and glm5.3.

@ai-sdk/openai-compatible is the generic adapter, the one that speaks to any API shaped like OpenAI’s. Do not use plain @ai-sdk/openai: that one expects the real OpenAI API.

Of everything inside each model, the only mandatory part is its key, which is the id. name, limit and modalities are optional: without them OpenCode works the same, you just see the bare id in the picker and automatic compaction runs on its own defaults. If you would rather have a short file, take them out.

glm5.3 will only answer if your key is on the premium tier. If it is not, leave it off the list so you do not pick it by mistake.

Where to put the key

In the example above the key is written inside the file, which is the most direct thing but not the most comfortable if that file ends up in a repository.

The alternative is to leave apiKey out of the opencode.json and store it with OpenCode’s /connect command, which writes it into its own credential store:

/connect

It will ask for a Provider id. Write exactly:

nan

The Provider id has to match the key in the JSON OpenCode pairs the credential with the provider by that identifier. If your provider is called nan in opencode.json and you write NaN, nan-builders or anything else in /connect, OpenCode saves the credential but associates it with no provider, and the requests go out with no key. It is the most common failure when connecting OpenCode to NaN, and it gives no message that explains it.

Then paste your API key and press Enter.

The context limits

"limit": { "context": 1048575, "output": 32768 }

limit.context and limit.output are the fields OpenCode reads. An older version of these docs published contextWindow, which is not part of OpenCode’s schema: an unknown key raises nothing anyone sees, OpenCode simply falls back to its own assumption about the window, and the symptom is a session that compacts far too early on the long-context models.

limit.context is the window the proxy accepts, which is not always the window the model was trained with: qwen3.8-flash is served at its native 262K, not at the YaRN-extended 1M. limit.output is a client-side budget rather than a server cap, so raise it if you need longer answers.

The compaction block

"compaction": { "auto": true, "prune": true, "reserved": 50000 }

OpenCode summarizes the conversation on its own when it gets close to the context limit, and reserved is the token margin it keeps to do that. With 1M token windows you are not going to touch the limit in a normal session, but leaving it on stops a very long session from being cut off abruptly.

Check that it works

opencode

Inside the agent, pick the model with /models and ask it for something short. If it answers, it is already going out through the cluster.

Recommended model

glm5.3-flash for code, deepseek-v4-flash for everything else. If you have the premium tier, glm5.3 for long sessions.