Pixeltable Integration Guide

This guide will walk you through integrating Pixeltable with APIpie, enabling you to build powerful multimodal AI applications with declarative data infrastructure and access to a wide range of language models.

What is Pixeltable?

Pixeltable is a declarative data infrastructure framework for multimodal AI apps that provides:

Data Ingestion & Storage: Work with images, videos, audio, documents, and structured data
Transformation & Processing: Apply Python functions or built-in operations automatically
AI Model Integration: Run inference (embeddings, object detection, LLMs) as part of data pipelines
Indexing & Retrieval: Create and manage vector indexes for semantic search
Incremental Computation: Only recompute what's necessary when data or code changes
Versioning & Lineage: Track data and schema changes for reproducibility

By connecting Pixeltable with APIpie, you gain access to a wide range of powerful language models for your multimodal applications while leveraging Pixeltable's sophisticated data management capabilities.

Integration Steps

1. Create an APIpie Account

Register here: APIpie Registration
Complete the sign-up process.

2. Add Credit

Add Credit: APIpie Subscription
Add credits to your account to enable API access.

3. Generate an API Key

API Key Management: APIpie API Keys
Create a new API key for use with Pixeltable.

4. Install Pixeltable

Install Pixeltable using pip:

pip install pixeltable

For additional functionalities, you might need extra packages:

# For working with embedding models
pip install sentence-transformers

# For object detection
pip install pixeltable-yolox

# For document processing
pip install spacy
python -m spacy download en_core_web_sm

5. Configure Pixeltable for APIpie

Pixeltable integrates with APIpie through built-in functions for OpenAI-compatible endpoints:

import pixeltable as pxt
from pixeltable.functions import openai

# Configure the OpenAI function with APIpie credentials
# This configuration can be reused across your tables
openai.configure(
    api_key="your-apipie-api-key",
    base_url="https://apipie.ai/v1"
)

Key Features

Unified Multimodal Interface: Work with images, videos, audio, documents, and structured data through a consistent interface
Declarative Computed Columns: Define transformations that run automatically on new or updated data
Built-in Vector Search: Add embedding indexes for similarity search directly on tables/views
On-the-Fly Data Views: Create virtual tables using iterators for efficient processing
Seamless AI Integration: Built-in functions for various AI providers including APIpie
Custom Python Functions: Extend with User-Defined Functions (UDFs)
Incremental Computation: Save time and costs by only recomputing what's necessary

Example Workflows

Application Type	What Pixeltable Helps You Build
Multimodal RAG Systems	Apps that search and generate content from diverse media
Computer Vision Apps	Applications performing detection and classification
Content Generation	Systems that create and modify text and images
Data Curation & Labeling	Tools for organizing and annotating multimodal datasets
AI Evaluation Frameworks	Systems to test and benchmark model performance

Using Pixeltable with APIpie

Basic Text Generation with APIpie

import pixeltable as pxt
from pixeltable.functions import openai

# Configure APIpie credentials
openai.configure(
    api_key="your-apipie-api-key",
    base_url="https://apipie.ai/v1"
)

# Create a table for prompts and responses
qa = pxt.create_table(
    'qa_system', 
    {'prompt': pxt.String}, 
    if_exists='replace'
)

# Add a computed column for LLM responses
qa.add_computed_column(
    response=openai.chat_completions(
        model='gpt-4o-mini',  # You can use any model available on APIpie
        messages=[{
            'role': 'user',
            'content': qa.prompt
        }]
    ).choices[0].message.content
)

# Insert a prompt and get a response
qa.insert([
    {'prompt': 'Explain quantum computing in simple terms.'}
])

# View the results
print(qa.select(qa.prompt, qa.response).collect())

Multimodal RAG with Pixeltable and APIpie

import pixeltable as pxt
from pixeltable.functions import openai, huggingface
from pixeltable.iterators import DocumentSplitter

# Configure APIpie credentials
openai.configure(
    api_key="your-apipie-api-key",
    base_url="https://apipie.ai/v1"
)

# Create a directory for organization
pxt.create_dir("rag_system", if_exists="replace")

# Create a document table and add some PDFs
docs = pxt.create_table(
    'rag_system.documents', 
    {'doc': pxt.Document},
    if_exists='replace'
)

docs.insert([
    {'doc': 'path/to/your/document.pdf'}
])

# Create chunks view with sentence-based splitting
chunks = pxt.create_view(
    'rag_system.chunks',
    docs,
    iterator=DocumentSplitter.create(
        document=docs.doc, 
        separators='sentence'
    )
)

# Add embedding index for similarity search
chunks.add_embedding_index(
    'text',
    string_embed=huggingface.sentence_transformer.using(
        model_id='all-MiniLM-L6-v2'
    )
)

# Define query function for retrieval
@pxt.query
def get_relevant_context(query_text: str, limit: int = 3):
    sim = chunks.text.similarity(query_text)
    return chunks.order_by(sim, asc=False).limit(limit).select(chunks.text)

# Create QA table that integrates context retrieval with APIpie
qa = pxt.create_table(
    'rag_system.qa', 
    {'question': pxt.String},
    if_exists='replace'
)

# Add context retrieval as a computed column
qa.add_computed_column(
    context=get_relevant_context(qa.question)
)

# Construct prompt with retrieved context
qa.add_computed_column(
    formatted_prompt=f"""
    Based on the following information, please answer the question.
    
    CONTEXT:
    {qa.context}
    
    QUESTION: {qa.question}
    
    ANSWER:
    """
)

# Generate response with APIpie
qa.add_computed_column(
    answer=openai.chat_completions(
        model='gpt-4o',
        messages=[{
            'role': 'user',
            'content': qa.formatted_prompt
        }]
    ).choices[0].message.content
)

# Ask a question
qa.insert([
    {'question': 'What are the key concepts discussed in the document?'}
])

# View the result
print(qa.select(qa.question, qa.answer).collect())

Image Classification with Pixeltable and APIpie Vision

import pixeltable as pxt
from pixeltable.functions import openai

# Configure APIpie credentials
openai.configure(
    api_key="your-apipie-api-key",
    base_url="https://apipie.ai/v1"
)

# Create an image table
images = pxt.create_table(
    'image_analysis', 
    {'image': pxt.Image}, 
    if_exists='replace'
)

# Insert some sample images
images.insert([
    {'image': 'https://upload.wikimedia.org/wikipedia/commons/thumb/6/68/Orange_tabby_cat_sitting_on_fallen_leaves-Hisashi-01A.jpg/1920px-Orange_tabby_cat_sitting_on_fallen_leaves-Hisashi-01A.jpg'},
    {'image': 'https://upload.wikimedia.org/wikipedia/commons/d/d5/Retriever_in_water.jpg'}
])

# Add a computed column for vision analysis using APIpie's vision model
images.add_computed_column(
    description=openai.chat_completions(
        model='gpt-4-vision',
        messages=[
            {
                'role': 'user',
                'content': [
                    {'type': 'text', 'text': 'Describe this image in detail.'},
                    {
                        'type': 'image_url',
                        'image_url': {'url': images.image}
                    }
                ]
            }
        ]
    ).choices[0].message.content
)

# View the results
print(images.select(images.image, images.description).collect())

Agentic Workflows with Tool Calling

import pixeltable as pxt
from pixeltable.functions import openai
import json

# Configure APIpie credentials
openai.configure(
    api_key="your-apipie-api-key",
    base_url="https://apipie.ai/v1"
)

# Define a UDF for weather information (mock function)
@pxt.udf
def get_weather(location: str) -> str:
    """Get current weather for a location."""
    # In a real app, you would call a weather API
    return f"The weather in {location} is sunny and 72°F."

# Define a UDF for search (mock function)
@pxt.udf
def search_web(query: str) -> str:
    """Search the web for information."""
    # In a real app, you would call a search API
    return f"Search results for: {query}..."

# Register the UDFs as tools
tools = pxt.tools(get_weather, search_web)

# Create agent table
agent = pxt.create_table(
    'agent_system', 
    {'user_request': pxt.String}, 
    if_exists='replace'
)

# Add a column for tool selection by the LLM
agent.add_computed_column(
    tool_choice=openai.chat_completions(
        model='gpt-4o',
        messages=[
            {'role': 'user', 'content': agent.user_request}
        ],
        tools=[t.to_openai_tool() for t in tools],
        tool_choice='auto'
    )
)

# Add a computed column to extract tool call information
@pxt.udf
def extract_tool_call(response) -> dict:
    """Extract tool call information from the LLM response."""
    tool_calls = response.choices[0].message.tool_calls
    if not tool_calls:
        return {"tool": None, "args": None}
    
    call = tool_calls[0]
    return {
        "tool": call.function.name,
        "args": json.loads(call.function.arguments)
    }

agent.add_computed_column(
    tool_info=extract_tool_call(agent.tool_choice)
)

# Add a column to invoke the selected tool
@pxt.udf
def invoke_tool(tool_info: dict) -> str:
    """Invoke the selected tool with the provided arguments."""
    if not tool_info or not tool_info["tool"]:
        return "No tool was selected."
    
    tool_name = tool_info["tool"]
    args = tool_info["args"]
    
    if tool_name == "get_weather":
        return get_weather(args["location"])
    elif tool_name == "search_web":
        return search_web(args["query"])
    
    return f"Unknown tool: {tool_name}"

agent.add_computed_column(
    tool_result=invoke_tool(agent.tool_info)
)

# Add a column for the final response
agent.add_computed_column(
    final_response=openai.chat_completions(
        model='gpt-4o-mini',
        messages=[
            {'role': 'user', 'content': agent.user_request},
            {'role': 'assistant', 'content': f"I found this information: {agent.tool_result}"}
        ]
    ).choices[0].message.content
)

# Test the agent
agent.insert([
    {'user_request': 'What\'s the weather like in Paris?'},
    {'user_request': 'Tell me about quantum computing.'}
])

# View the results
print(agent.select(agent.user_request, agent.tool_info, agent.final_response).collect())

Troubleshooting & FAQ

How do I update computed columns when data changes?
Pixeltable automatically recomputes values when data or UDFs change. To force a recomputation, use table.invalidate_columns(['column_name']).
How do I handle environment variables securely?
Store your API keys in environment variables and load them in your code. Pixeltable also supports configuration files for security.
Can I use APIpie's routing capabilities with Pixeltable?
Yes, APIpie's routing features work seamlessly with Pixeltable. Simply specify the desired model when calling functions.
How do I monitor costs and usage?
Pixeltable's incremental computation helps minimize API calls, but you should still monitor your APIpie dashboard for usage metrics.
Can I cache API calls to reduce costs?
Yes, Pixeltable's computed columns implicitly cache results, only recomputing when inputs change.
How can I scale to large datasets?
Pixeltable is designed to handle large datasets efficiently. For very large datasets, consider using Views with iterators for on-demand processing.

For more information, see the Pixeltable documentation or the GitHub repository.

Support

If you encounter any issues during the integration process, please reach out on APIpie Discord or Pixeltable Discord for assistance.

What is Pixeltable?​

Integration Steps​

1. Create an APIpie Account​

2. Add Credit​

3. Generate an API Key​

4. Install Pixeltable​

5. Configure Pixeltable for APIpie​

Key Features​

Example Workflows​

Using Pixeltable with APIpie​

Basic Text Generation with APIpie​

Multimodal RAG with Pixeltable and APIpie​

Image Classification with Pixeltable and APIpie Vision​

Agentic Workflows with Tool Calling​

Troubleshooting & FAQ​

Support​