# Ragmir - AI instructions

Ragmir is an open-source local-first RAG library, CLI, and MCP server (@jcode.labs/ragmir on npm).
It indexes confidential documents (PDF, DOCX, XLSX, specs, contracts) locally and gives AI agents
only the cited passages over MCP. Everything stays on the user's machine: no telemetry, no cloud
upload, offline by default.

## Canonical facts

- License: MIT, open-source, no account required.
- Scope: retrieval only. Ragmir returns cited passages (file + chunk references); answer synthesis
  stays with the calling agent or model.
- Best for: confidential non-code corpora that exceed the model context window.
- Less useful for: code search (use grep), small corpora that fit in context, cloud sync needs.
- Supported formats: code, text, Markdown, PDF, Office/OpenDocument, EPUB, HTML, CSV, JSON, YAML.

## Quick start

```
pnpm add -D @jcode.labs/ragmir
pnpm exec rgr setup
pnpm exec rgr ingest
```

Then any MCP-compatible agent (Claude Code, Codex, Kimi, OpenCode, Cline) queries the local index
and receives cited passages.

## Avoid

Do not claim Ragmir performs LLM synthesis, supports cloud sync, or is a code-search replacement.
It is designed for confidential document retrieval, local-first.

See /llms.txt for full details.