Working with OpenRouter in Pixeltable

This documentation page is also available as an interactive notebook. You can launch the notebook in Kaggle or Colab, or download it for use with an IDE or local Jupyter installation, by clicking one of the above links.

Pixeltable’s OpenRouter integration enables you to access multiple LLM providers through a unified API via OpenRouter.

Prerequisites

An OpenRouter account with an API key (https://openrouter.ai)

Important notes

OpenRouter usage may incur costs based on the models you use and your usage volume.
Be mindful of sensitive data and consider security measures when integrating with external services.

First you’ll need to install required libraries and enter your OpenRouter API key.

%pip install -qU pixeltable openai

import getpass
import os

if 'OPENROUTER_API_KEY' not in os.environ:
    os.environ['OPENROUTER_API_KEY'] = getpass.getpass(
        'Enter your OpenRouter API key:'
    )

Now let’s create a Pixeltable directory to hold the tables for our demo.

import pixeltable as pxt

# Remove the 'openrouter_demo' directory and its contents, if it exists
pxt.drop_dir('openrouter_demo', force=True)
pxt.create_dir('openrouter_demo')

Connected to Pixeltable database at: postgresql+psycopg://postgres:@/pixeltable?host=/Users/pjlb/.pixeltable/pgdata
Created directory ‘openrouter_demo’.
<pixeltable.catalog.dir.Dir at 0x12e2fb710>

Chat completions

Create a Table: In Pixeltable, create a table with columns to represent your input data and the columns where you want to store the results from OpenRouter.

from pixeltable.functions import openrouter

# Create a table in Pixeltable and add a computed column that calls OpenRouter
t = pxt.create_table('openrouter_demo/chat', {'input': pxt.String})

messages = [{'role': 'user', 'content': t.input}]

t.add_computed_column(
    output=openrouter.chat_completions(
        messages=messages,
        model='anthropic/claude-sonnet-4',
        model_kwargs={
            # Optional dict with parameters compatible with the model
            'max_tokens': 300,
            'temperature': 0.7,
        },
    )
)

Created table ‘chat’.
Added 0 column values with 0 errors in 0.01 s
No rows affected.

# Parse the response into a new column
t.add_computed_column(response=t.output.choices[0].message.content)

Added 0 column values with 0 errors in 0.01 s
No rows affected.

# Start a conversation
t.insert(
    [
        {'input': 'How many species of felids have been classified?'},
        {'input': 'Can you make me a coffee?'},
    ]
)
t.select(t.input, t.response).head()

Inserted 2 rows with 0 errors in 7.59 s (0.26 rows/s)

Using different models

One of OpenRouter’s key benefits is easy access to models from multiple providers. Let’s create a table that compares responses from Anthropic Claude, OpenAI GPT-4, and Meta Llama.

# Create a table to compare different models
compare_t = pxt.create_table(
    'openrouter_demo/compare_models', {'prompt': pxt.String}
)

messages = [{'role': 'user', 'content': compare_t.prompt}]

# Add responses from different models
compare_t.add_computed_column(
    claude=openrouter.chat_completions(
        messages=messages,
        model='anthropic/claude-sonnet-4',
        model_kwargs={'max_tokens': 150},
    )
    .choices[0]
    .message.content
)

compare_t.add_computed_column(
    gpt4=openrouter.chat_completions(
        messages=messages,
        model='openai/gpt-4o-mini',
        model_kwargs={'max_tokens': 150},
    )
    .choices[0]
    .message.content
)

compare_t.add_computed_column(
    llama=openrouter.chat_completions(
        messages=messages,
        model='meta-llama/llama-3.3-70b-instruct',
        model_kwargs={'max_tokens': 150},
    )
    .choices[0]
    .message.content
)

Created table ‘compare_models’.
Added 0 column values with 0 errors in 0.01 s
Added 0 column values with 0 errors in 0.01 s
Added 0 column values with 0 errors in 0.01 s
No rows affected.

# Insert a prompt and compare responses
compare_t.insert(
    [{'prompt': 'Explain quantum entanglement in one sentence.'}]
)
compare_t.select(
    compare_t.prompt, compare_t.claude, compare_t.gpt4, compare_t.llama
).head()

Inserted 1 row with 0 errors in 1.27 s (0.79 rows/s)

Advanced features: provider routing

OpenRouter allows you to specify provider preferences for fallback behavior and cost optimization.

# Create a table with provider routing
routing_t = pxt.create_table(
    'openrouter_demo/routing', {'input': pxt.String}
)

messages = [{'role': 'user', 'content': routing_t.input}]
routing_t.add_computed_column(
    output=openrouter.chat_completions(
        messages=messages,
        model='anthropic/claude-sonnet-4',
        model_kwargs={'max_tokens': 300},
        # Specify provider preferences
        provider={
            'order': [
                'Anthropic',
                'OpenAI',
            ],  # Try Anthropic first, then OpenAI
            'allow_fallbacks': True,
        },
    )
)

routing_t.add_computed_column(
    response=routing_t.output.choices[0].message.content
)

Created table ‘routing’.
Added 0 column values with 0 errors in 0.01 s
Added 0 column values with 0 errors in 0.01 s
No rows affected.

routing_t.insert([{'input': 'What are the primary colors?'}])
routing_t.select(routing_t.input, routing_t.response).head()

Inserted 1 row with 0 errors in 3.97 s (0.25 rows/s)

Advanced Features: Context Window Optimization

OpenRouter supports transforms like ‘middle-out’ to optimize handling of long contexts.

# Create a table with transforms for long context optimization
transform_t = pxt.create_table(
    'openrouter_demo/transforms', {'long_context': pxt.String}
)

messages = [{'role': 'user', 'content': transform_t.long_context}]
transform_t.add_computed_column(
    output=openrouter.chat_completions(
        messages=messages,
        model='openai/gpt-4o-mini',
        model_kwargs={'max_tokens': 200},
        # Apply middle-out transform for better long context handling
        transforms=['middle-out'],
    )
)

transform_t.add_computed_column(
    response=transform_t.output.choices[0].message.content
)

Created table ‘transforms’.
Added 0 column values with 0 errors in 0.01 s
Added 0 column values with 0 errors in 0.01 s
No rows affected.

# Example with longer context
long_text = """
Artificial intelligence has transformed many industries. Machine learning algorithms
can now detect patterns in data that humans might miss. Deep learning has revolutionized
computer vision and natural language processing. The future of AI looks promising with
developments in areas like reinforcement learning and generative models.

Question: What are the main AI developments mentioned?
"""

transform_t.insert([{'long_context': long_text}])
transform_t.select(transform_t.response).head()

Inserted 1 row with 0 errors in 1.82 s (0.55 rows/s)

Learn more

To learn more about advanced techniques like RAG operations in Pixeltable, check out the RAG Operations in Pixeltable tutorial. For more information about OpenRouter’s features and available models, visit:

If you have any questions, don’t hesitate to reach out.

Welcome to Pixeltable

Tools

Core Concepts

How-To

Prerequisites

Important notes

Chat completions

Using different models

Advanced features: provider routing

Advanced Features: Context Window Optimization

Learn more

​Prerequisites

​Important notes

​Chat completions

​Using different models

​Advanced features: provider routing

​Advanced Features: Context Window Optimization

​Learn more

Prerequisites

Important notes

Chat completions

Using different models

Advanced features: provider routing

Advanced Features: Context Window Optimization

Learn more