Pinecone
85Managed vector database for similarity search over billions of embeddings, with serverless and BYOC options
3 products our research names as a replacement for MongoDB Atlas in Vector Databases & RAG, each with what it does differently — prices, API and status from our own checks rather than from the vendors’ own pages.
Managed vector database for similarity search over billions of embeddings, with serverless and BYOC options
Open-source vector search engine written in Rust, available self-hosted or as a managed cloud for RAG and recommendation apps.
Open-source Postgres backend with auth, storage, realtime, edge functions and a pgvector toolkit for AI apps.
Not named as direct replacements — the neighbours in the same part of the catalog, best-rated first.
Turns documents and past sessions into a graph-plus-vector memory that AI agents can query across runs.
Open-source search engine with vector storage and hybrid keyword-plus-semantic retrieval, self-hosted or run as a cloud service
Postgres backend bundled with a RAG pipeline, agent runtime and a visual workflow builder
Builds an open-source, Postgres-based memory engine that gives AI agents relational, vector, and graph retrieval.
Open-source backend platform (auth, database, storage, functions) increasingly pitched as the data layer under AI agents and RAG apps.
Source-available multimodal retrieval engine that embeds whole document pages, so RAG works over diagrams and datasheets.
Open-source search and vector engine that combines text, vectors, ranking and model inference for RAG, recommendations and agents.
Social data API and hosted MCP server for X, Instagram, TikTok and Reddit without platform API keys
Meta's family of open-weight LLMs, downloadable under a community licence with a 700M-MAU carve-out.
Runs ComfyUI workflows in a browser on rented GPUs, billing separately for edit time and generation time.
EU-hosted LLM gateway with an OpenAI-compatible API and routing by price, quality or latency.
A routing and verification layer that sits between your apps and language models to cut cost and catch ungrounded answers.
Not what you were after? Browse all Vector Databases & RAG →