beam_knowledge

beam_knowledge ingests text into a vector store and retrieves the closest chunks for a query.

beam_knowledge = { path = "../riverbeam/crates/beam_knowledge", features = ["postgres"] }

Feature	Effect
postgres	PgVectorStore, ensure_knowledge_schema
polars-ingest	Ingest statistics computed with Polars

Types

Item	Role
Embedder	Trait: name, dimensions, embed, embed_query
HashEmbedder, MockEmbedder	Hermetic embedders for development and tests
OpenAiCompatEmbedder::openai(..), ::nvidia(..)	HTTP embedders (POST {base_url}/embeddings)
VectorStore	Trait: upsert, query, delete, replace_document_chunks
InMemoryVectorStore, PgVectorStore	Cosine similarity in memory or with pgvector
Chunker, SimpleChunker	Paragraph packing, 800 characters by default
Retriever, DenseRetriever	Embed the query and search the store
KnowledgeBuilder	Chunk, embed and upsert; with_chunker
KnowledgeManager	ingest, delete_document, retrieve
RagContextStrategy	Render hits as a context block
build_knowledge, build_knowledge_from_env, KnowledgeBuildRequest, EmbedderKind, StoreKind	The factory
KnowledgeScope, RetrievalFilter, Hit, Chunk, VectorRecord, Namespace	Data types
IngestionResult, IngestStats, ingest_stats	Ingest outcome and statistics

The environment variables, the fallback rules and the current limitations are in Knowledge.