Pi.

Pi is a deliberately small terminal agent: a minimal shell that adapts to how you work instead of imposing how it works. It speaks the OpenAI format, so NaN goes in as one more provider.

You can skip all of this

The NaN CLI writes this configuration for you: Setup tab, e to paste your key, space to mark Pi, c to apply.

Configuration

Declare the provider

In ~/.pi/agent/models.json:

{
  "providers": {
    "nan": {
      "baseUrl": "https://api.nan.builders/v1",
      "api": "openai-completions",
      "apiKey": "sk-your-key",
      "compat": { "supportsDeveloperRole": true },
      "models": [
        {
          "id": "glm5.3-flash",
          "name": "GLM 5.3 Flash",
          "reasoning": true,
          "input": ["text", "image"],
          "contextWindow": 1000000,
          "maxTokens": 16384
        },
        {
          "id": "deepseek-v4-flash",
          "name": "DeepSeek V4 Flash",
          "reasoning": true,
          "input": ["text", "image"],
          "contextWindow": 1000000,
          "maxTokens": 16384
        }
      ]
    }
  }
}

api: "openai-completions" is what tells Pi which format to speak. maxTokens is the ceiling for the answer on each request, not the context.

Make it the default

In ~/.pi/agent/settings.json:

{
  "defaultProvider": "nan",
  "defaultModel": "glm5.3-flash"
}

This step is not optional

Without defaultProvider and defaultModel, Pi keeps using its factory provider and what you get is a 401. It is the most common failure when connecting it, and the error message does not say this file is the problem.

Try it

pi

Ask it for something short. If it answers, it is going out through the cluster. You can switch models mid-session without leaving the agent.

Recommended model

glm5.3-flash for coding work. deepseek-v4-flash for everything else, which also has the widest quota on the cluster.

Known issues

The key is written in the file. models.json lives in your home directory, so it does not usually end up in a repository, but keep it in mind if you sync your configuration between machines.
maxTokens is not the context. It is the ceiling for each answer. Reasoning comes out of that same budget, so if you ask for long reasoned answers, raise it.
The models you declare are the ones you get. Pi does not ask the cluster what is available: it shows whatever is on the list.