pi-env-var-provider

A pi extension that registers a custom OpenAI-compatible provider from environment variables. Configure baseUrl, apiKey, and model settings without editing models.json — perfect for proxies, API gateways, and local models.

Quick Start

# Set required environment variables
export PI_BASE_URL=https://proxy.example.com/v1
export PI_API_KEY=sk-your-api-key

# Optional: customize provider/model names
export PI_PROVIDER_NAME=my-proxy
export PI_MODEL_ID=gpt-4o

# Load and run with pi
pi -e npm:@mcowger/pi-env-var-provider

# Then select the model
/model my-proxy/gpt-4o

Two Modes of Operation

Mode 1: New Provider (Default)

Register a completely new provider with custom models:

export PI_BASE_URL=https://api.example.com/v1
export PI_API_KEY=sk-...
export PI_PROVIDER_NAME=custom-ai      # Provider identifier
export PI_MODEL_ID=llama-70b           # Model ID
export PI_MODEL_NAME="Llama 3 70B"     # Display name
pi -e npm:@mcowger/pi-env-var-provider

Mode 2: Override Existing Provider

Change the baseUrl for a built-in provider (keeps all models):

export PI_BASE_URL=https://my-gateway.openai.azure.com
export PI_API_KEY=...
export PI_OVERRIDE_PROVIDER=openai     # Override built-in "openai" provider
pi -e npm:@mcowger/pi-env-var-provider

All openai/* models now use your custom baseUrl.

Environment Variables

| Variable | Required | Default | Description | |----------|----------|---------|-------------| | PI_BASE_URL | Yes | — | API base URL (e.g., https://api.openai.com/v1) | | PI_API_KEY | Mode 1 only | — | API key or access token | | PI_OVERRIDE_PROVIDER | No | — | Existing provider to override instead of creating new | | PI_PROVIDER_NAME | No | env-provider | New provider identifier | | PI_MODEL_ID | No | default | Model ID | | PI_MODEL_NAME | No | matches PI_MODEL_ID | Display name | | PI_MODEL_REASONING | No | false | Supports reasoning/thinking | | PI_MODEL_INPUT | No | text,image | Input types (text, image, or both) | | PI_CONTEXT_WINDOW | No | 128000 | Context window size in tokens | | PI_MAX_TOKENS | No | 16384 | Max output tokens | | PI_COST_INPUT | No | 0 | Input cost per 1M tokens | | PI_COST_OUTPUT | No | 0 | Output cost per 1M tokens | | PI_COMPAT_OVERRIDES | No | — | OpenAI compat overrides as JSON |

Common Use Cases

Azure OpenAI

export PI_BASE_URL=https://my-resource.openai.azure.com/openai/deployments/my-deployment
export PI_API_KEY=...
export PI_OVERRIDE_PROVIDER=openai
export PI_MODEL_ID=gpt-4o
export PI_COMPAT_OVERRIDES='{"supportsReasoningEffort":false}'
pi -e npm:@mcowger/pi-env-var-provider

Local Ollama

export PI_BASE_URL=http://localhost:11434/v1
export PI_API_KEY=ollama  # or any non-empty string
export PI_PROVIDER_NAME=ollama-local
export PI_MODEL_ID=llama3.1
export PI_CONTEXT_WINDOW=8192
pi -e npm:@mcowger/pi-env-var-provider

# Then: /model ollama-local/llama3.1

OpenRouter Proxy

export PI_BASE_URL=https://openrouter.ai/api/v1
export PI_API_KEY=sk-or-...
export PI_OVERRIDE_PROVIDER=openai
export PI_MODEL_ID=claude-sonnet-4
pi -e npm:@mcowger/pi-env-var-provider

Perplexity API

export PI_BASE_URL=https://api.perplexity.ai
export PI_API_KEY=pplx-...
export PI_PROVIDER_NAME=perplexity
export PI_MODEL_ID=sonar-pro
export PI_MAX_TOKENS=8192
pi -e npm:@mcowger/pi-env-var-provider

GitHub Action Usage

Use with shaftoe/pi-coding-agent-action for CI/CD workflows:

- name: Run Pi agent with custom provider
  uses: shaftoe/pi-coding-agent-action@v2
  with:
    github_token: ${{ secrets.GITHUB_TOKEN }}
    provider: env-provider        # or PI_OVERRIDE_PROVIDER value
    model: default                # or PI_MODEL_ID value
    token: ${{ secrets.API_KEY }} # or inline in PI_API_KEY
    extensions: npm:@mcowger/pi-env-var-provider
  env:
    PI_BASE_URL: ${{ secrets.CUSTOM_API_URL }}
    PI_API_KEY: ${{ secrets.API_KEY }}
    PI_PROVIDER_NAME: my-gateway
    PI_MODEL_ID: gpt-4o

GitHub Action with Provider Override

- name: Run Pi with Azure OpenAI
  uses: shaftoe/pi-coding-agent-action@v2
  with:
    github_token: ${{ secrets.GITHUB_TOKEN }}
    provider: openai
    model: gpt-4o
    token: ${{ secrets.AZURE_OPENAI_KEY }}
    extensions: npm:@mcowger/pi-env-var-provider
  env:
    PI_BASE_URL: ${{ secrets.AZURE_OPENAI_ENDPOINT }}
    PI_API_KEY: ${{ secrets.AZURE_OPENAI_KEY }}
    PI_OVERRIDE_PROVIDER: openai

Advanced: Compat Overrides

For providers with OpenAPI compatibility quirks, use PI_COMPAT_OVERRIDES:

export PI_COMPAT_OVERRIDES='{
  "supportsDeveloperRole": false,
  "supportsReasoningEffort": false,
  "maxTokensField": "max_tokens",
  "requiresToolResultName": true
}'

Available options:

• supportsDeveloperRole — Use "system" instead of "developer" role
• supportsReasoningEffort — Provider supports reasoning_effort parameter
• maxTokensField — Use "max_tokens" instead of "max_completion_tokens"
• requiresToolResultName — Tool results must include name field
• supportsUsageInStreaming — Include usage in streaming responses
• thinkingFormat — Format for thinking blocks ("openai", "zai", "qwen")

Installation

Global (all projects)

pi install npm:@mcowger/pi-env-var-provider

Project-local

pi install -l npm:@mcowger/pi-env-var-provider

Temporary (one-time)

pi -e npm:@mcowger/pi-env-var-provider

From Git (latest)

pi install git:github.com:mcowger/pi-env-var-provider

Requirements

• pi (any recent version)
• @mariozechner/pi-coding-agent (bundled with pi)

Uninstalling

pi remove npm:@mcowger/pi-env-var-provider

License

MIT © mcowger