Pi API setup: configuration and troubleshooting

Pi is an open-source terminal coding agent with a small core that grows through extensions. Custom endpoints go in `models.json`; no extension is needed.

Updated:

Set up FlyRouter

Set the API key

Terminal
export FLYROUTER_API_KEY="sk-..."

Add a provider

Edit ~/.pi/agent/models.json, creating it if it does not exist:

~/.pi/agent/models.json
{
  "$schema": "https://pi.dev/schemas/models.schema.json",
  "providers": {
    "flyrouter": {
      "baseUrl": "https://api.flyrouter.ai/v1",
      "api": "openai-completions",
      "apiKey": "$FLYROUTER_API_KEY",
      "models": [
        { "id": "claude-sonnet-5-5", "name": "Claude Sonnet 5.5", "input": ["text", "image"] },
        { "id": "gpt-6.1-sol", "name": "GPT-6.1 Sol", "input": ["text", "image"] }
      ]
    }
  }
}

An apiKey written as $NAME is read from the environment; a literal key also works. The id in models is the model ID sent to the API.

Choose a model

Type /model in Pi and pick a FlyRouter model. The file is re-read every time /model opens, so no restart is needed after an edit.

Model parameters

Without them, Pi assumes a 128,000-token context window and 16,384 output tokens. If a model's real limits differ, add contextWindow and maxTokens to its entry, and "reasoning": true for models that support thinking.

Troubleshooting

  • 404: baseUrl must include /v1.
  • 401: the environment variable is not set, or the name in apiKey is wrong.
  • Cost shows $0.00: Pi estimates cost from the cost block in models.json and shows 0 without one. The FlyRouter console usage records are what you are actually charged.