Collections Overview
Build production-grade Retrieval-Augmented Generation (RAG) applications without managing external Pinecone, Qdrant, or Milvus clusters. The Collections API provides automated document chunking, dense vector embeddings, and sub-50ms hybrid semantic search.
What are Collections?
Managed Vector Storage
High-performance HNSW index graph with automated rebalancing and encrypted per-tenant partitions.
Automated Chunking
Intelligent AST and semantic boundary parsing for PDFs, markdown, code, and spreadsheets.
Automated chunking and embedding
When documents are added to a collection, Plannix splits content into overlapping chunks (default 512 tokens) and computes high-dimensional vector representations using plannix-embed-v1.
Available collection endpoints
Hybrid lexical + semantic retrieval
Plannix combines BM25 keyword matching with dense cosine distance reranking, ensuring both exact technical identifiers and conceptual synonyms are surfaced with high precision.
Multi-tenant vector isolation
Vectors are encrypted at rest with workspace-specific KMS keys. Data is strictly isolated and never accessible by queries outside your authenticated organization.