Could not load provider 'azure-openai'

[paulbatum]

I'm unable to configure the Azure OpenAI provider using amplifier provider use azure-openai. The command fails with:

Error: Could not load provider 'azure-openai'
Configuration cancelled.

Other providers (anthropic, openai) work fine.

Steps to Reproduce:

1. Install amplifier: uv tool install git+https://github.com/microsoft/amplifier
2. Run: amplifier init (successfully installs all providers including azure-openai)
3. Run: amplifier provider list (shows azure-openai in the list)
4. Run: amplifier provider use azure-openai
5. Error occurs immediately before any configuration prompts

Expected Behavior:

The command should prompt for Azure OpenAI configuration (endpoint, deployment, credentials) similar to how other providers work.

Actual Behavior:

The command fails immediately with "Could not load provider 'azure-openai'" before showing any prompts.

Root Cause

please note that the following analysis was generated by claude code and has not been independently verified:

The issue is in how the Azure OpenAI provider validates credentials during init():

amplifier-module-provider-azure-openai/init.py, lines 138-141

if api_key is None and token_provider is None:
raise ValueError("api_key or token_provider must be provided")

client = AsyncOpenAI(base_url=base_url, api_key=api_key or token_provider)

When amplifier provider use calls get_provider_info() to discover configuration fields, it uses _try_instantiate_provider() with placeholder credentials (empty string for api_key). The Azure OpenAI provider:

1. Accepts the empty string (passes the None check)
2. But then the underlying AsyncOpenAI client rejects the empty api_key with: "The api_key client option must be set either by passing api_key to the client or by setting the OPENAI_API_KEY environment variable"

This causes get_provider_info() to return None, which causes the "Could not load provider" error.

Why Other Providers Work:

The anthropic, openai, and ollama providers don't validate credentials or create clients during init() - they defer client creation until first use. This allows them to be instantiated for discovery purposes (calling get_info()) without valid credentials.

Suggested Fix:

The Azure OpenAI provider should allow instantiation without valid credentials, similar to other providers. Options:

1. Lazy client initialization: Don't create the AsyncOpenAI client in init(), create it on first use
2. Make credentials optional for info/discovery: Add a flag or separate method that allows getting provider info without credentials
3. Accept placeholder credentials: Modify the validation to allow placeholder/empty values when the provider is being instantiated for discovery only

Environment:

• amplifier version: Latest from main branch (installed via uv tool install git+https://github.com/microsoft/amplifier)
• Python: 3.12.11
• OS: Linux (WSL2)
• All providers cached in ~/.amplifier/module-cache/ with latest commits

[natalie-isak]

+1, experiencing the same issue!

[paulbatum]

I think i found a workaround - set a dummy openai key env var:
export OPENAI_API_KEY="placeholder"

That should prevent the error and allow you to configure the provider.