Open-source RAG that cites its sources and won't lock your models

What it does

Agentset is an open-source platform for building and shipping retrieval-augmented generation apps. It handles the full lifecycle: document ingestion, chunking, embedding, vector indexing, retrieval, and a chat playground where answers include citations. It also offers hosted deployments with custom domains and a developer API with typed SDKs.

The interesting bit

Most RAG tools force you into a specific model or database; Agentset stays model-agnostic and throws in built-in multi-tenancy, which is usually the part you end up building yourself. It is essentially a full-stack application built on Next.js, Prisma, Supabase, and Trigger.dev, offered as both a managed cloud service and a self-hosted repo.

Key highlights

• Model-agnostic: bring your own LLM, embeddings, and vector store
• Chat playground with message editing and source citations
• Built-in multi-tenancy, not bolted on later
• Typed SDKs and an OpenAPI spec for API access
• Available as managed cloud or self-hosted

Caveats

• The README mentions evaluation and benchmarks but doesn’t explain what they measure or how to use them.
• Ingestion is described as “turnkey,” yet supported file formats and parsing behavior are left to the docs.

Verdict

Teams that need a turnkey RAG backend with tenant isolation and model flexibility should look here. If you already have a custom retrieval pipeline you love, this might be more framework than you need.