Skip to content
Karajan Code

kaRAGan — karajan-rag

karajan-rag (kaRAGan) is the context engine of the Karajan family: index a codebase, documents or data with one command and decide — with data, not faith — which context travels to the model. Everything runs through the same guarded path: sensitivity policy plus PII redaction on every output towards an LLM.

Terminal window
npm install -g karajan-rag
karajan-rag index ./my-project
karajan-rag query "how is billing calculated?" ./my-project

A context-strategy engine

Not just retrieval: rag (top-k chunks), cag (whole corpus, cache-friendly) and hybrid (whole files picked by retrieval) over the same index — and eval --compare-modes to choose with numbers.

Sensitivity first

public / internal / confidential per corpus or path prefix. The policy routes every LLM call; redactPII runs in depth. Audited end to end — the full third-party review trail is published verbatim.

Serve it anywhere

MCP server for agents, HTTP API with a self-contained web playground, or embed it with createRag() — same RagService underneath, from laptop to Cloud Run.

No silent fallbacks

Missing peer, invalid config or incompatible vector space fail loudly with the exact fix. The local flow is deterministic: no credentials, no network.

Start here

Section titled “Start here”
RAG in 5 minutes Index, query, answer — offline and credential-free.
Context strategies rag vs cag vs hybrid, and how to decide with data.
Serve and configure MCP for agents, HTTP + playground, project defaults.
Sensitivity and privacy Levels, policy routing and PII redaction.
Container and GCP Docker and Terraform: private by default on Cloud Run.
Embeddable SDK createRag() in Astro, Next, Fastify or any Node worker.