pi-volcengine-provider

Pi provider extension for Volcengine Ark Coding API models.

Package details

extension

Install pi-volcengine-provider from npm and Pi will load the resources declared by the package manifest.

$ pi install npm:pi-volcengine-provider

Pi manifest JSON

{
  "extensions": [
    "./index.ts"
  ]
}

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

Pi package that registers a Volcengine Ark Coding API provider.

After this package is published to npm:

pi install npm:pi-volcengine-provider

You can also install directly from GitHub:

pi install git:github.com/MacroSony/pi-volcengine-provider

For local development:

pi -e /path/to/pi-volcengine-provider

The provider id is volcengine-plan. Pi will use ~/.pi/agent/auth.json first, then the VOLCENGINE_API_KEY environment variable as a fallback.

{
  "volcengine-plan": { "type": "api_key", "key": "your-volcengine-key" }
}

Or:

export VOLCENGINE_API_KEY=your-volcengine-key

This extension registers:

Run pi --list-models after installing to confirm the provider is loaded.

This extension maps Pi's native thinking selector to provider-specific Chat Completions controls:

Pi off -> thinking: { "type": "disabled" }
Most Volcengine reasoning models: Pi low, medium, high -> thinking: { "type": "enabled" } plus matching reasoning_effort
Pi minimal is hidden because Volcengine's reasoning_effort: "minimal" means no thinking; selecting it clamps to low
Pi xhigh -> reasoning_effort: "max" only for models verified to accept it (ark-code-latest, deepseek-v4-flash, deepseek-v4-pro)
glm-5.1 uses the GLM/Z.AI thinking shape: Pi low, medium, and high send thinking: { "type": "enabled" } without reasoning_effort; Pi off sends disabled.