@vendoai/knowledge

@vendoai/knowledge is Vendo's product knowledge base: the concrete retrieval engines and the ingestion pipeline that sit behind core's frozen KnowledgeAdapter contract (@vendoai/core).

One contract, three engines:

• Built-in local engine — lexical retrieval over the host's own store (the free tier), reading and writing the vendo_knowledge_docs / vendo_knowledge_chunks collections.
• Cloud client — speaks the knowledge wire protocol (vendo/knowledge-wire@1) to a managed backend.
• BYO HTTP template — the same wire protocol pointed at a host-supplied endpoint.

Ingestion (parse → normalize → structural chunk → sync) turns a host's local sources into document-level upserts; chunking, embedding, and indexing belong to the engine behind the contract.

The umbrella package wires the resolved adapter into the agent loop so knowledge-backed retrieval grounds chat, generated apps, and action planning with citations, under the same guard policy and audit trail as every other tool.

Status

Stage 0 scaffold (ENG-355). The package + toolchain are in place; the engines and ingestion land across Stages 1–3 of the knowledge build. See the @vendoai/knowledge project in Linear.