Skip to content
jCode

Open sourceDeveloper tool

Home page of ragmir.com: the title “Agentic RAG. For developers. With your sources.”, install and use case buttons, and a terminal demo where a coding agent adds project documents to Ragmir
ragmir.com, October 2026

Open sourceDeveloper tool

Ragmir: open-source agentic RAG, with your sources

Ragmir gives coding agents such as Claude Code and Codex cited passages from your code and documents. It indexes them on your machine and points each answer to its line, page or slide.

  • Open source
  • MCP server
  • CLI and TypeScript API
  • Local index
Visit ragmir.comSource code on GitHub
Licence
AGPL-3.0-only from v3 · Commercial option
Version
6.3.1, 27 October 2026
npm downloads
1,259 a week (npm page, 1 November 2026)
Interfaces
CLI · TypeScript API · MCP
001
The tool

Cited context for your coding agents

Your agent asks a question; Ragmir answers with the exact passages from your repository, its documentation and office files, and where each one comes from, so every claim can be checked.

Indexing and retrieval run locally. A third-party service you connect receives the passages it is given, under its own settings and data policy.

How it works

  1. Index your project

    Choose the code, docs and office files: Ragmir builds a local LanceDB index that pairs BM25 with optional embeddings, and keeps it up to date incrementally.

  2. Connect your agents

    One command sets up Claude Code and Codex, which query Ragmir through read-only MCP tools.

  3. Ask in plain words

    Your agent, or you in the terminal, gets the passages that answer the question, lexical and vector results fused by reciprocal rank, even offline.

  4. Check the source

    Every passage points to its line, page or slide, so you can verify it before you rely on it.

002
Screenshots

Ragmir on GitHub, npm and its website

Public GitHub repository jcode-works/jcode-ragmir: the project files, 303 commits, the AGPL-3.0 licence and topics such as MCP server, local RAG and TypeScript
GitHub repository
TypeScript source of Ragmir's lexical scoring on GitHub: the BM25 functions that weight term frequency and document length
BM25 scoring code
npm page of the @jcode.labs/ragmir package, version 6.3.1: install command, AGPL-3.0-only licence, 1,259 weekly downloads and the start of the README
npm package
Use cases section of ragmir.com: four workflows, and a diagram where Claude Code or Codex searches the project documents with Ragmir before changing code and tests
Use cases, ragmir.com
003
What sets it apart

Cited passages from code and documents, on your machine

The passages a coding agent needs, and where each one comes from, without a hosted index or a bundled chat model.

  1. Citations you can check

    Every passage carries its native coordinates: lines in code, pages in a PDF, slides, spreadsheet cells or EPUB sections. Ragmir also flags a source that changed since it was indexed, and evidence too thin to rely on.

    • Lines, pages, slides, cells
    • Changed-source flag
    • Weak-evidence flag

  2. Code and office documents in one index

    Your code and your documents share one local index, refreshed as files change. A local OCR engine reads scanned PDF pages, and only those that have no text.

    • Code · Markdown · JSON
    • PDF · Word · Excel · PowerPoint
    • OpenDocument · EPUB
    • Local OCR
    • Incremental indexing

  3. Hybrid search that works offline

    BM25 and vector results are fused by reciprocal rank, aware of file paths, camelCase names and code definitions, in a LanceDB index. The default mode downloads no model; multilingual embeddings run locally when you turn them on.

    • BM25 and vectors
    • LanceDB
    • Optional local embeddings

  4. Made for coding agents

    Four read-only MCP tools, scoped to one project. On the project's benchmark of 27 October 2026, the right file was in the top ten for 93% of 101 questions on a 43,874-chunk monorepo, with a 40 ms median search.

    • MCP, read-only tools
    • Claude Code and Codex
    • No telemetry
004
Technologies

How Ragmir is built

An open-source TypeScript package for Node.js, published on npm, whose code anyone can read on GitHub.

Search

  • BM25 lexical search
  • Hybrid BM25 and vector search
  • LanceDB local index
  • Transformers.js embeddings, optional
  • Offline by default

Agents

  • MCP server
  • MCP SDK
  • Claude Code
  • Codex
  • Compact citations

Documents

  • Code and Markdown
  • PDF pages
  • Office documents
  • EPUB
  • Citations by file, page and segment

Privacy and quality

  • Sensitive-file exclusions
  • Local index
  • No telemetry
  • Security audit command
  • Type checking
  • Tests
  • AGPL-3.0 · commercial option
005
Explore more

Related products and services

Other productWorkoutGenMCP toolingBuild MCP tools for your coding agentsAI featuresAdd AI features to your product
006
Contact

Let's talk about your project.

Tell me what you want to build or improve, your budget and your target date.

  • Fully remote
  • English or French
  • France · UK · US
Describe your project on MaltAsk a question on LinkedInDownload my CV
007
FAQ

Questions about Ragmir

What stays local?

Indexing and retrieval. A connected client can send the passages it receives to an external service, so confidentiality also depends on that client and its settings.

How can I integrate Ragmir?

Through its command line, its TypeScript API or its MCP server. The repository documents the requirements and supported configurations.

Which licence applies?

AGPL-3.0-only from version 3, with a separate commercial licence available. Earlier releases keep their published licence.

Who develops Ragmir?

I do, as founder of jCode. I created Ragmir in 2026: its library, CLI and MCP server, with automated tests and benchmarks.

Can jCode help my team set it up?

Yes. Technical consulting can cover choosing the documents to index and connecting Ragmir to your coding agents, in your repositories.