Pixeltable

Pixeltable is an open-source multimodal AI data infrastructure product for developers building applications that need to store media, run models, index embeddings, serve endpoints, and version data. It presents these pieces as unified in one declarative Python table, aiming to replace the need to stitch together multiple services and glue code.

Its table system treats images, video, audio, and documents as first-class citizens, alongside JSON, arrays, binary, string, and float values. The page also describes built-in versioning and time travel, lineage to every cell, and computed columns that recompute only affected cells when a model or user-defined function changes. For retrieval, it includes embedding indexes with vector search that stay synchronized as data and models change. It also offers views and iterators for tasks such as chunking documents, extracting video frames, splitting audio, and slicing strings. In addition, Pixeltable lists UDFs, built-in transforms, and first-party AI providers in one layer, with batching, retries, rate limits, and caching handled for those integrations. The providers named include OpenAI, Anthropic, Gemini, Bedrock, Mistral, Groq, DeepSeek, Hugging Face, CLIP, Whisper, Voyage, Jina, Together, Fireworks, Ollama, Replicate, fal.ai, RunwayML, BFL, FLUX, and Twelve Labs.

The same page says tables can be used as agents, with UDFs and @pxt.query as tools, persistent memory, and reasoning traces. It also says tables and queries can become HTTP endpoints through FastAPIRouter and pxt serve, with background jobs included. Export is described in the form of SQL, Parquet, PyTorch, and more. Delivery is via pip install pixeltable, and the site states that it is free and open source under Apache 2.0 with no account required.