A LOCAL RAG PIPELINE · MCP-NATIVE · MIT

Solid ground
for AI-native software.

Ragolith indexes your code and documentation into a local knowledge base, then serves them to any MCP-aware LLM. So your agents can ship faster, modernize legacy with confidence, and unify scattered knowledge.

Get started View on GitHub

— BUILT FOR THE MODERN SDLC

Three reasons it earns a place in your stack.

Accelerate the SDLC with AI

LLMs guess less when they have the right context. Ragolith feeds your agents precise code, docs, and design rationale — so they generate, review, and refactor with confidence.

Modernize legacy, fearlessly

Semantic search across decades-old codebases surfaces dependencies, patterns, and risk before a single line changes. Migrate, port, and rewrite — guided by what's actually there.

Consolidate under one umbrella

Code, PDFs, Word docs, ADRs, SQL — chunked, embedded, and searchable under one roof. End the documentation diaspora.

— HOW IT WORKS

Three layers. Zero external APIs.

Everything runs locally in Docker. Your code never leaves your machine.

1

Ingest

Clone repos. Walk files respecting .gitignore. Read PDF / DOCX. Dispatch to language-aware chunkers. Batch into Weaviate.
2

Search

Hybrid BM25 + vector retrieval. Cross-encoder rerank. Autocut at the largest score gap. Diversity filter. Tuned for code, not prose.
3

Serve

MCP server over stdio exposes ten tools — search, find symbol, file structure, callers, callees — to Claude Desktop, IDEs, or any MCP client.

— THE BEDROCK

What you get out of the box.

Local-first. Embeddings and reranking run in Docker. No API keys, no data leaving your perimeter.
AST-aware chunking. TypeScript, JavaScript, Java, C#, SQL — split at function and class boundaries so results are self-contained.
Symbols & call graphs. Find callers, callees, and definitions across the whole codebase.
Incremental ingest. git diff since last SHA — re-index only what changed.
Hybrid search. BM25 catches exact identifiers, vectors catch semantic intent.
S3-backed backups. Filesystem snapshots with optional S3 push / pull.

— QUICK START

From clone to query in four steps.

# 1. install
npm install

# 2. start Weaviate + embedding + reranker
npm run weaviate:up

# 3. configure your projects
cp ragc.config.example.json ragc.config.json

# 4. ingest
npm run ingest

Then wire ragolith-server into your MCP client and ask your agent anything about the codebase.

Plant a flag on your codebase.

Open source. MIT licensed. Built for engineering teams shipping with AI.

Star on GitHub

Solid ground for AI-native software.