#PI

A minimal, extensible open-source terminal coding agent (by earendil-works). By default the model gets just four tools — read / write / edit / bash — extended on demand via Extensions, Skills and Pi Packages. A custom provider is one JSON file away.

#Install

npm install -g --ignore-scripts @earendil-works/pi-coding-agent

npx @earendil-works/pi-coding-agent

yay -S pi-coding-agent-bin

npm package: @earendil-works/pi-coding-agent; website: pi.dev.

#Prerequisites

• PI installed (pi --version prints a version)
• An OpenAI-format group key from the API Keys page

#Configure

Edit ~/.pi/agent/models.json and add ByteSeek as a custom provider:

{
  "providers": {
    "byteseek": {
      "baseUrl": "https://tokens.byteseek.ai/v1",
      "api": "openai-completions",
      "apiKey": "sk-your-key",
      "models": [
        { "id": "gpt-5.6-sol", "name": "GPT-5.6 Sol", "reasoning": true },
        { "id": "gpt-5.6-terra", "name": "GPT-5.6 Terra", "reasoning": true }
      ]
    }
  }
}

• Model ids come from the model plaza; prefix them when using a composite key.
• apiKey accepts the environment form "$BYTESEEK_API_KEY" if you prefer not to store the key on disk.
• For an Anthropic-format group key, set api to anthropic-messages and baseUrl to https://tokens.byteseek.ai (no /v1).

The file reloads automatically every time /model opens — no session restart after edits.

#Verify

Start pi, switch via /model (or Ctrl+L) to a byteseek model, and send a message. Charges appear on the usage page.

#Common issues