Pixeltable is a declarative data platform that treats video, audio, images, documents and embeddings as first‑class column types in versioned tables, preserving schema, lineage and time‑travel capabilities. It automatically runs AI models as computed columns and maintains synchronized embedding indexes, enabling incremental pipelines, vector search, and seamless integration with annotation tools for multimodal machine‑learning workflows.
Funding
$5.5M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
7OTGFounders
Product
Problem
Organizations building multimodal AI applications often store video, audio, images, and documents in disparate locations and rely on custom scripts to move, process, and track this data. This fragmentation makes it difficult to maintain data integrity, version control pipelines, and keep AI‑generated annotations in sync with source assets.
Solution
Pixeltable offers a declarative, persistent data platform that treats multimodal assets as first‑class column types within versioned tables. Users define schemas that include Image, Video, Audio, Document, and embedding arrays, enabling automatic lineage, schema evolution, and time‑travel queries without manual glue code. Computed columns let AI models run as part of the table definition, so model outputs (e.g., object detections, embeddings) are materialized and kept up to date whenever new rows are inserted or existing data changes. Built‑in embedding indexes provide declarative vector search that stays synchronized with the underlying data. The system also supports incremental pipelines, snapshotting, and integration with popular tools such as Label Studio and FiftyOne, allowing teams to iterate on data preparation, annotation, and curation efficiently.
Target Audience
Primary users are ML engineers, data scientists, and AI product teams that need a reliable, versioned data layer for building video, image, audio, and document‑centric machine‑learning pipelines.
Features
- Multimodal column types (Image, Video, Audio, Document, Array, Json) that store files by reference while preserving schema, versioning, and lineage
- Declarative computed columns that execute AI models (e.g., HuggingFace, OpenAI) automatically on data insertion or update
- Automatic version control and time‑travel queries for tables, enabling reproducible experiments and easy rollback
- Embedding indexes with configurable similarity metrics that stay in sync with data and support fast vector search
- Snapshot and view mechanisms for point‑in‑time data freezes and reusable query definitions
- Native integrations for annotation workflows (Label Studio, FiftyOne) and export pipelines
- Open‑source Python SDK with support for cloud storage backends (S3, GCS, Azure) and minimal dependency footprint