Framer AI AgentsDesign and publish professional sites with AI
Promoted
VectraSDK
Open Source provider-agnostic RAG pipeline for production AI
9 followers
Open Source provider-agnostic RAG pipeline for production AI
9 followers
VectraSDK is an end-to-end RAG pipeline SDK for building fast, accurate, and production-ready AI applications. It handles ingestion, chunking, embeddings, vector storage, retrieval, reranking, memory, and generation, all in one modular SDK. Provider-agnostic by design, it supports multiple LLMs, embedding models, and vector databases across Python and Node, so you can ship grounded AI without vendor lock-in or boilerplate.
Hey Product Hunt 👋
Vectra exists because building RAG in the real world is still way harder than it should be.
Teams aren’t struggling with prompts they’re struggling with ingestion pipelines, retrieval quality, provider lock-in, and staying flexible as models and vector databases keep changing. On top of that, it’s often hard to even see what’s happening inside a RAG system once it’s running.
Most tools today sit at one of two extremes:
Very low-level building blocks that leave you wiring everything together, or
Heavy frameworks that lock you into specific providers, patterns, assumptions or its just complex to use.
Once you commit to an embedding model, vector DB, or LLM, changing your mind later often means refactoring half your system. That makes experimentation expensive and iteration slow.
Vectra is an open-source, provider-agnostic RAG platform that gives you a complete context pipeline out of the box:
Ingestion
Chunking
Embeddings
Vector storage
Retrieval
Reranking
Memory
Generation
Observability and evaluation
All with production-readiness.
You can swap LLMs, embedding models, or vector databases without touching your application code, while still using advanced retrieval strategies like Hybrid search, HyDE, and LLM-based reranking.
Vectra also includes built-in observability, so you can understand what’s happening at each stage of the pipeline what was retrieved, how it was reranked, where latency comes from, and how changes affect quality instead of treating RAG as a black box.
The goal is simple:
👉 Let teams focus on building their AI product, not rebuilding RAG infrastructure.
👉 Let the SDK handle orchestration and visibility not your code.
This launch is just the beginning, and I’d love your feedback:
What are you building with RAG?
What’s missing today?
What would make RAG feel even more boring (in a good way)?
Thanks for checking it out 🙌, your support matters a lot, if you find this helpfull provide an upvote and star the github repos
https://github.com/iamabhishek-n/vectra-js
https://github.com/iamabhishek-n/vectra-py