Open-source RAG infrastructure that survives production workloads. Upload documents, query via API, get answers with sources. Hybrid search, multimodal, complex reasoning - all included. Model-agnostic. Used by 1,500+ teams in medical AI, legal tech, enterprise search.

Hey Product Hunt 👋

Abdellatif here, founder of Agentset.

The story: Last year we needed RAG for 9M pages. Followed tutorials, got a prototype in a week. Tested on 100 docs - looked great.

Deployed to production - completely failed.

We spent 3 months debugging everything - query generation, reranking, custom chunking, metadata injection. My co-founder and I have built products for 5M+ users, but RAG humbled us hard.

After processing 5M+ documents across two enterprises, we finally cracked it. We thought: nobody else should waste 3 months like this.

So we built Agentset - open-source RAG that works in production, out of the box. All the painful lessons baked in: agentic reasoning, hybrid search + reranking, automatic citations, multimodal support.

Built with some amazing tools:

1,500+ teams are using it now. Feels incredible seeing others skip the pain we went through.

Try it: https://agentset.ai

Happy to answer anything about RAG, our journey, or what we learned in the trenches!

— Abdellatif

The honest story about 3 months debugging RAG in production is what makes this credible. Does Agentset expose hooks for custom chunk boundaries, or does the parser handle semantic splits by default? That's usually where production accuracy lives or dies.

Hi everyone!
So I basically built a fraud detector which has it's backend using gemini to generate NL explanations over flagged transactions. But the retrieval accuracy over the structured financial data was basically where it broke down.

Does Agentset exposes custom chunking hooks? For like tabular/graphical data cause in my opinion chunking strategy matters more than the model choice.