
AI is empowering everyone to create data-driven documents. We're proud to announce our @OpenAI ChatGPT plugin integrating Noteable's interactive notebook platform into ChatGPT. Ready for the details? Buckle up for an exciting ride! 🧵

Acquired by @confluentinc. Data Collaboration was hard. We made it easier.

AI is empowering everyone to create data-driven documents. We're proud to announce our @OpenAI ChatGPT plugin integrating Noteable's interactive notebook platform into ChatGPT. Ready for the details? Buckle up for an exciting ride! 🧵

🚀 @pixeltablehq + @pydantic integration! Type safety in Pixeltable just got even better! Define your data models once, get validation everywhere across your storage and orchestration layers: -> pip install pixeltable pydantic | @samuelcolvin

An alternative to Text-to-SQL We launched pxt.retrieval_udf(), a feature that transforms any Pixeltable table into an AI-queryable tool for agentic workflows. Traditional RAG excels at unstructured data (at least Pixeltable does) but struggles with structured information. Now your AI agents can query databases, CRM systems, and knowledge bases using natural language while maintaining data precision. Entire companies are being built just to solve this problem, It's a UDF in Pixeltable. Let's start Pixeltabuling.

Pixeltable: AI Data Infrastructure Providing a Declarative, Incremental Approach for Multimodal Apps

Good luck! You gonna need a Data Infra that actually handles videos/images/audio then! @pixeltablehq is built by the team who created parquet (the file format), Twitter, Amazon, and Airbnb ML Platforms... You'd be surprise how much we abstract data plumbing for you.

If you want to help build the next-gen open source multimodal AI data infrastructure, DMs open. You'll work alongside people who built: F1/BigTable, parquet format, Impala, Yellowbrick, MapR, Dremio, SanDisk and built the ML platforms at Airbnb/Twitter (X), & Amazon.


The Ultimate Python Library for Multimodal AI Meet Pixeltable—a unified, declarative framework that handles your entire multimodal pipeline, from data storage to model execution. Seamlessly manage images, videos, text, and tabular data—all in one place. 100% open-source.

Memory is simply storage, orchestration, retrieval and indexing. No need for fancy abstraction. You just need a multimodal data infrastructure: github.com/pixeltable/pixelt… What matters is the pre- and post-processing from the given “MCP” client to maintain that external memory equivalent to a user bringing their own USB flash drives to each computer they are working on. Btw if you need a multimodal open source infra at Box - happy to chat! Just pip install.

We should find a more efficient way to configure LLMs other than wasting 24k tokens.

Thanks for the post @svpino. The updated location for this tutorial is: github.com/pixeltable/pixela…

Pixelagent isn't just another agent framework. It's a collection of clear, concise examples demonstrating how to build robust, customizable agent functionality directly on top of @pixeltablehq's data infrastructure. I hope it inspires people to build their own agent class from scratch, fork Pixelagent to build their own framework, and build the abstraction they want based on their use cases.

Here is a step-by-step tutorial on building a complete stateful OpenAI agent. (If you know Python, this is all you need to learn to build a working agent in an hour, including complete explanations of every component.) This agent: • It's conversational • Has persistent memory • Can use external tools This tutorial uses Pixelagent, a fully open-source agent engineering blueprint. Pixelagent is at the intersection of an LLM, storage, and orchestration. You can use it with the following features: • Supports automated data orchestration • Multimodal with support for text, images, audio, etc. • Handles OpenAI and Anthropic message protocols • Supports tool integrations • Supports reasoning, reflection, memory, knowledge, and team workflows Pixelagent is built on top of Pixeltable's declarative framework, which unifies data storage, retrieval, and orchestration. I'm pasting a link below to show an end-to-end live agent built on top of this framework.

What do you use for coding in Cursor, as of today?

Happy to chat. Long-term memory here in 20 lines of code, lazy evaluation, efficient caching/parallelization... all in the database: docs.pixeltable.com/docs/exa… I've a couple of sample apps on Github where I'm using it already. An infinite-memory discord and multimodal local chatbot.

Try in Carmel.
I feel like an old man saying this, but—$27M sure doesn't go as far as it used to.