Proposal: Adding Microsoft Fabric Integration for Azure OpenAI

[pawarbi]

Hi Pixeltable team!

I'm interested in contributing a new integration for Microsoft Fabric that would allow Pixeltable users to seamlessly access Azure OpenAI services within the Fabric ecosystem. I wanted to start a discussion before submitting a PR to get your feedback on the approach. I am part of Microsoft's product group.

I would like to add Fabric as a client in Pixeltable so our customers & users can use it seamlessly on our platform. Below is my proposal:

The integration would add a Fabric client that handles:

Automatic endpoint discovery via get_fabric_env_config()
Token-based authentication via TokenUtils
Support for Azure OpenAI chat completions (and potentially embeddings)

Proposed scope
For the initial PR, I'm thinking:

• Fabric OpenAI client (pixeltable/ext/functions/fabric. py)
• chat_completions() function for chat models
• embeddings() function for embedding models
• Automatic authentication handling
• Unit tests in tests/ext/functions/
• Tutorial notebook demonstrating usage within Fabric
• Documentation updates

Basic chat completion will look like this

import pixeltable as pxt
from pixeltable.ext.functions import fabric

# Create a table for customer support tickets
pxt.drop_table('support. tickets', ignore_errors=True)
pxt.create_dir('support', ignore_errors=True)

tickets = pxt.create_table('support.tickets', {
    'ticket_id': pxt.Int,
    'customer_message': pxt.String,
    'priority': pxt.String
})

# Add a computed column that automatically generates AI responses
# No API keys needed - Fabric handles authentication! 
tickets.add_computed_column(
    ai_response=fabric.chat_completions(
        messages=[
            {"role": "system", "content":  "You are a helpful customer support agent.  Be concise and professional."},
            {"role": "user", "content":  tickets.customer_message}
        ],
        model="gpt-5"
    )
)

# Insert data - AI responses are generated automatically
tickets.insert([
    {'ticket_id': 1, 'customer_message':  'How do I reset my password?', 'priority':  'low'},
    {'ticket_id': 2, 'customer_message': 'My order hasnt arrived after 2 weeks', 'priority': 'high'},
    {'ticket_id': 3, 'customer_message':  'Can I change my subscription plan?', 'priority': 'medium'}
])

# Query results with AI-generated responses
tickets.select(
    tickets.ticket_id,
    tickets.customer_message,
    tickets.ai_response
).collect()

My Questions:

• Does this integration align with Pixeltable's roadmap and goals?
• Should this be structured similarly to the existing OpenAI integration, or as a separate module?
• Any preferences on how to handle the Fabric-specific dependencies (synapse-ml-fabric)? I can add mock tests.
• Any other guidance to successfully integrate in your library?

[aaron-siegel]

Hi @pawarbi ,

Yes, we'd definitely welcome a PR for Fabric integration. Thank you!

Note that we already have Azure OpenAI working in Pixeltable: it is officially supported via the existing OpenAI SDK integration with appropriate configuration parameters. If you look at test_azure_openai in tests/test_openai.py you can see an example where we call an Azure model deployment from Pixeltable.

I'd welcome more context on how Fabric integration will differ from vanilla Azure OpenAI integration - my own understanding is somewhat weak in this area. If it's wire compatible with OpenAI via the OpenAI SDK, I'm not sure it needs a separate functions module; perhaps all we need is a way to leverage the Fabric configuration/authentication in Pixeltable. If the SDK is different, then a separate module is warranted. The ext package has been sunsetted, so any new modules can go into functions.

Of the other items you mention: authentication handling, unit tests, tutorial notebook, and docs all make sense to include either way.

Your other questions:

• Does this integration align with Pixeltable's roadmap and goals?

Yes!

• Any preferences on how to handle the Fabric-specific dependencies (synapse-ml-fabric)? I can add mock tests.

Add them to our dev dependencies in pyproject.toml. This will make them part of our dev environment (which is fine) while keeping them out of the base Pixeltable install. No need for mock tests: all our integrations are tested live.

• Any other guidance to successfully integrate in your library?

Read CONTRIBUTING.md if you haven't already and see if you can successfully run make tests and make check. You may get some concurrency errors with the former (this is normal) and have to run pytest -v --lf afterwards to re-run the failed tests.

Thank you! Feel free to hop on our discord and ping me if you want to chat about any follow ups:
https://discord.com/invite/QPyqFYx2UN

Aaron

[pawarbi]

Thank you @aaron-siegel . Microsoft Fabric is a SaaS platform in Azure data ecosystem. Although Fabric uses Azure OpenAI, it's deployment endpoints are auto-provisioned and authenticated for each customer which means the Azure OpenAI client will not work is. I will look into both the options to be safe and share it before creating the PR.
Appreciate your response.

[pawarbi]

PR submitted #1109 (comment)