Declarative Data Infrastructure•For AI Applications

How Pixeltable Works

Understand the core concepts that make building AI applications seamless.

Declarative Pipelines

Define entire AI workflows, from data transformation to model inference (like embeddings or LLM calls) and tool integration, using simple Python functions as computed columns. Pixeltable automatically manages dependencies, orchestrates execution, and computes results incrementally, simplifying the creation of complex agents and RAG systems.

Learn More

1. Python Code (Define Pipeline)

import pixeltable as pxt
import pixeltable.functions as ptf

# Table with image URLs
t = pxt.create_table('images', {'url': pxt.String})
t.insert([
  {'url': 'http://example.com/image1.jpg'},
  {'url': 'http://example.com/image2.png'}
])

# Resize images (computed column)
t.add_computed_column(
    resized_img=ptf.pil.resize(t.url, size=(128, 128))
)

# Access resized images
results = t.select(t.url, t.resized_img).collect()

2. Pixeltable Table (Computed Result)

Table after resizing images

url	resized_img
image1.jpg	(128x128 Image)
image2.png	(128x128 Image)

Unified Multimodal Datastore

Natively manage diverse data types – images, videos, audio, documents, embeddings, and structured data – within the same table structure. Ingest new data or reference existing files (local, S3) without duplication, eliminating the need for separate vector DBs, file stores, or feature stores when building multimodal AI applications.

Learn More

Example Pixeltable Table

ID	Media	Source	AI Output (Text)	AI Output (Vector)
1		/data/images/cat.jpg	A curious cat looking directly at the camera...	[0.12, -0.45, ...]
2		s3://my-bucket/videos/beach.mp4	Drone footage flying over a sunny beach...	[-0.73, 0.22, ...]
3	Review: Great product, fast shipping!	api://reviews/12345	Positive sentiment analysis.	[0.91, 0.01, ...]
4		/docs/financial_report_q3.pdf	Key insights extracted from Q3 financial report via LLM...	[0.33, -0.11, ...]

Automatic Incremental Updates

Computed columns and views automatically update when source data or function code changes. Only recompute what's necessary – if new documents are added, only their embeddings and derived data are calculated, saving significant time and resources.

Learn More

System Idle. Existing data processed.

Source Data / Events

Pixeltable (Auto-Updated View)

ID	Data	Computed
1	data_a.jpg	(Processed: v1)
2	data_b.png	(Processed: v1)

Pixeltable automatically updates only the necessary data when changes occur.

Built-in Vector Search & RAG

Create embedding indexes effortlessly on text, images, video frames, and more. Perform semantic search co-located with metadata, retrieve relevant context, optionally augment with tool outputs, and build faster, more reliable Retrieval-Augmented Generation (RAG) applications.

Learn More

Simulated Image Search

Try: , ,

Enter a search term above or click an example.

Find relevant images based on your query.

Ready to build your AI application?

Start building with Pixeltable today and transform how you work with multimodal data.

Get Started View Documentation