[RaGBlocks]

RAG is broken.

And not in a “still early” kind of way — it's broken at the foundation.

If you've built serious retrieval systems, you already know this.

Even if no one wants to admit it.

Another wrapper won’t fix it.

Neither will another LangChain abstraction.

This is a call to stop pretending brittle pipelines are production-ready AI infrastructure.

We're building the modular RAG infrastructure we wish we had

Modular. Agentic. Versioned. Self-evolving. Easily deployable.

If you’ve built real RAG systems, you have felt this pain

• Ingestion logic that breaks silently across Notion, Drive, Confluence, S3
• Chunking that loses semantic structure or bloats your context window
• No version control — no idea what your retriever is actually seeing
• Every query routed to a generic LLM instead of SQL, vector, or plugin logic
• Zero memory. No feedback loops. No metrics. Just vibes.

This is where the magic ends. And where the maintenance begins.

That is why we are building RAGBlocks

RAGBlocks is an open-core framework to build real-world retrieval systems from first principles — composable, observable, versioned. Not just demos. Not just pipelines. Systems.

Modular blocks — retrievers, SQL agents, memory, rerankers, evaluators — all isolated
Declarative flows — testable, inspectable, and diffable across versions
Infrastructure-portable — cloud, hybrid, or air-gapped - no lock-in
Evaluation is built-in — not a Jupyter afterthought

We're not building for everyone.

We're looking for 2-3 founding partners who already feel this pain. Consultants, CTOs, or infra leads who are tired of rebuilding the same broken stack and are ready to invest in a better way forward.

APPLY TO PARTNER