We built Elysia for you: An open-source, agentic RAG app

Hello Weaviate Community! 🤗

We're making it easier than ever to bring agentic AI to your data. This week, we’ve launched Elysia, an open-source framework that spins up an end-to-end agentic RAG application with just two terminal commands.

We’ll also explore the nuances of fine-tuning embedding models and share the latest AI research on GEPA from the Weaviate podcast with Connor.

Let's dive in!

Getting started is ridiculously simple:

pip install elysia-ai 
elysia start

🔗 GitHub | 📖 Demo | 🎮 Blog post

Latest AI & tech insights

Explore our recent Weaviate content:

Read

• ✍️ Why, When, and How to Fine-Tune a Custom Embedding Model: Key considerations for customizing an embedding model through fine-tuning it on company- or domain-specific data to improve the downstream retrieval performance in RAG applications. Read the blog

Watch

• 📽️ GEPA Explained! GEPA introduces a breakthrough in DSPy by showing that Reflective Prompt Evolution—powered by LLM-driven prompt writing, mutation, and system-aware optimization—can outperform cutting-edge RL methods like GRPO, enabling more effective and adaptive AI systems even with training at test-time. Watch the video

• 🎙️GEPA with Lakshya A. Agrawal - Weaviate Podcast #127! Lakshya A. Agrawal, a U.C. Berkeley Ph.D. student, led GEPA’s innovations in using LLMs as optimizers, discussed in the podcast with topics like Test-Time Training and LangProBe benchmarks. Watch the podcast

• 📺 Short explainers on key concepts:

  • Simple Chunking Techniques

  • Advanced Chunking Techniques

  • Late Chunking

Product highlights

Check out our latest product updates. Or follow along in the #announcement channel in the Weaviate Community Slack.

• Weaviate Core v1.32.3 – Now with OpenAI GPT-5 model support, improved vector dimensions validation, async replication fixes, and enhanced queue retries. Read more »

• Model2Vec Inference v1.1.0 – Added two new static embedding models: minishlab/potion-retrieval-32M (English, retrieval-focused) and minishlab/potion-multilingual-128M (multilingual, distilled BGE-M3). Read more »

• Transformers Inference v1.12.1 – New dimensions parameter lets you control vector size for Matroyshka models. Read more »

• Weaviate Helm v17.5.1 – Added support for text2vec-model2vec and usage metrics for GCS & S3 modules. Read more »

Want to leverage these new features without managing infrastructure? Deploy a cluster in minutes on Weaviate Cloud.

Weaviate recipes

• 🎉 GPT-5 just dropped. Try it now

• 💻 Build a 100% local RAG pipeline with OpenAI’s GPT-OSS and Weaviate! Go to the recipe

Upcoming events

Join us online or in person for expert-led AI workshops and networking events. Connect, learn, and build with the community!

Online

• 🛠️ Intro to building AI-native applications with Weaviate | August 19. Register here

• 📚 Weaviate Office Hours: Ask us anything | August 20. Register here

• 📚 Weaviate Office Hours: Ask us anything | August 27. Register here

• ⚡ Vector Databases: What the Heck?! [JavaScript Edition] | August 28. Register here

In-person

• 🌆 San Francisco | AWS AI Hack Day Micro Conference | August 22. Save your spot

• 🌇 San Francisco | 2025 AI Conference Hack Day in San Francisco | September 16. Save your spot

• 🌉 San Francisco | Hack Night at GitHub | September 17. Save your spot

• 🤖 San Francisco | Robotics & AI Meet | September 17. Save your spot

Company updates

We are excited to welcome Dhruv, who is joining us from Boston, MA 🇺🇸 as a Partnerships Intern.

Hungry for more?

Have a question or want to connect? Say hi in the Community Slack or join our Weaviate Forum to engage in community conversations.

Follow our Blog, LinkedIn, Twitter, and YouTube for more content updates.

Weaviate is open source ➡️ Check our GitHub and star us while you're there. ⭐

See you in two weeks,

Femke