Skip to contents

securecontext 0.2.0

New public API

  • vector_store$metadata(id) and vector_store$ids() accessors for safely inspecting store contents without reaching into R6 privates.
  • knowledge_store$get_metadata(key) accessor for retrieving the metadata list attached to a key.
  • embed_custom(fn, dims, name) — generic adapter that wraps any function producing n x dims embedding matrices as a securecontext_embedder.
  • embed_openai(model, dims, api_key, ...) — convenience embedder that calls the OpenAI /v1/embeddings endpoint via httr2 (Suggests).

Fixes

  • context_for_chat() no longer reaches into R6 private fields (..__enclos_env__$private$.metadata) to pull chunk text; it now uses the new public vector_store$metadata() accessor.

Notes on embedding coverage

  • ellmer does not currently expose embeddings, so there is no embed_ellmer(). Build on embed_custom() if you need to route through a different SDK.

securecontext 0.1.0

  • Initial CRAN release.
  • S7-based securecontext_document class for representing text documents with metadata.
  • chunk_text() for splitting documents into overlapping chunks by token count.
  • TF-IDF embedder (embed_tfidf()) for local, dependency-free text embeddings.
  • vector_store R6 class with cosine similarity search and RDS persistence.
  • knowledge_store R6 class for persistent JSONL key-value storage.
  • context_builder for token-aware, priority-based context assembly.
  • count_tokens() utility using word-based approximation.
  • Optional integration with ‘ellmer’ and ‘orchestr’ for LLM agent workflows.