Vector Streaming is popular for its low latency and high modularity features. But the best of it all is the memory leak that we are trying to bring it but haven’t been able to click with the audience yet. Why we named it vector streaming and how it plays a major role in memory management.
After presenting at Google, PyCon DE, Berlin Buzzwords, and GDG Berlin, I was surprised by how many people approached me with questions about writing configurations, chunk sizes, and batch sizes for EmbedAnything. Since I had never specifically covered this topic in my talks or blog posts, I decided to create this comprehensive guide to clarify these concepts and explain how we handle your chunking strategy with vector streaming.
Super Excited to share the latest development in our library, which essentially giving you more embedding choices -- Cohere and siglip, new chunking method-- late chunking and more crates that facilitates amazing modality and maintainability for our rust codebase, --processor crate. so let's dive in.
Lumo, a powerful Rust-based agent, offers seamless integration with MCPs (Modular Control Protocols) and remarkable flexibility in implementation. While Lumo can be used as a library, CLI tool, or server, this guide will focus specifically on deploying Lumo in server mode for optimal MCP integration.
In the rapidly evolving landscape of AI agents, particularly those employing Large Language Models (LLMs), observability and tracing have emerged as fundamental requirements rather than optional features. As agents become more complex and handle increasingly critical tasks, understanding their inner workings, debugging issues, and establishing accountability becomes paramount.
The 2025 PyCon DE event highlighted a growing but cautious interest in AI agents among the Python community. While agent technology received significant attention, many speakers and attendees expressed skepticism about their practical utility in real-world applications.
We have received multiple requests to add different vector database for our vector streaming. So We have decided to put a detailed guide for different vector databases out there. We are happy to accept pull-request.
When working with domain-specific queries, we often struggle with the challenge of balancing in-domain and out-of-domain requests. But not anymore! With embedanything, you can leverage fine-tuned, domain-focused models while smolagent takes the lead in smart decision-making. Whether you're handling queries from different domains or need to combine their insights seamlessly, smolagent ensures smooth collaboration, merging responses for a unified, accurate answer.
We are thrilled to share that EmbedAnything version 0.5 is out now and comprise of insane development like support for ModernBert and ReRanker models. Along with Ingestion pipeline support for DocX, and HTML let’s get in details.
The best of all have been support for late-interaction model, both ColPali and ColBERT on onnx.
ColPali, a late-interaction vision model, leverages this power to enable text searches within images. This means you can pinpoint the exact pages in a PDF containing relevant text, even if the text exists only as part of an image. For example, suppose you have hundreds of pages in a PDF and even hundreds of PDFs. In that case, ColPali can identify the specific pages matching a query—an impressive feat for streamlining information retrieval. This system is widely come to be known as Vision RAG.