pi-requesty (Official Requesty extension for Pi)

The official Requesty extension for the Pi Coding Agent

(Recommended) Install from Github Repo

pi install git:github.com/requestyai/pi-requesty

To get the latest version later, run pi update --extensions, then restart Pi or run /reload.

To stay on a fixed version, append a full 40 character commit SHA (@<sha>). pi update fails on short SHAs and never moves a pinned install.

Install locally

Check out the code from the official code repository https://github.com/requestyai/pi-requesty, and then:

pi install ./pi-requesty

To run once without installing:

pi -e ./pi-requesty

Configuration

The extension resolves the requesty key from the same two sources pi falls back through, in this order:

1. a stored api_key credential in the agent directory's auth.json
2. providers.requesty.apiKey in models.json

(A runtime --api-key is not consulted: model discovery calls <baseUrl>/models directly, outside pi's request pipeline. A stored credential that cannot be resolved — a failing command, an unset variable — falls through to models.json rather than failing, which is friendlier than pi's own behaviour for inference.)

Both accept a literal value, a $VARIABLE reference, or a !command that prints the key — so the key can stay in your secret manager:

{
  "requesty": {
    "type": "api_key",
    "key": "!secret-tool lookup requesty api-key"
  }
}

Only the provider settings are read from ~/.pi/agent/models.json; when the file does not exist yet, defaults are used (https://router.requesty.ai/v1 with the openai-completions API).

The provider always sends HTTP-Referer: https://pi.dev and X-Title: Pi, which turn on Requesty's automatic prompt caching for Pi and label the traffic as Pi in analytics. Headers set in models.json take precedence, and /requesty-models-sync writes the defaults into the file when they are missing.

A !command runs with the same shell pi uses (on Windows, pi's configured shell, usually Git Bash). It runs twice at startup: once for model discovery and once when pi resolves the key for inference.

Example using a literal key instead:

{
  "providers": {
    "requesty": {
      "name": "Requesty",
      "baseUrl": "https://router.requesty.ai/v1",
      "apiKey": "rqsty-sk-...",
      "api": "openai-completions",
      "models": []
    }
  }
}

On startup, the extension fetches both <baseUrl>/models and <baseUrl>/models/managed using the resolved key as the bearer token, merges the two lists (deduplicated by model ID), and registers the discovered models with pi. Each request is bounded by a 10s timeout, because it runs before pi finishes loading. Without a credential discovery is skipped silently: pi resolves credentials for inference itself, and the models already listed in models.json keep working. The key is never written to models.json.

Command

Inside pi:

/requesty-models-sync

The command resolves the key as described above, fetches the Requesty models and writes the discovered model IDs back to ~/.pi/agent/models.json. The file is rewritten owner-only (0600), because it can hold a literal key. Run /reload after syncing.

Tests

npm test