georgeguimaraes/arcana

Embeddable RAG library for Elixir/Phoenix with agentic pipelines and dashboard

What it solves

Arcana is an embeddable RAG (Retrieval-Augmented Generation) library for Elixir and Phoenix applications. It eliminates the need for separate vector databases, indexing services, or orchestration layers by integrating directly into an existing Ecto repository (PostgreSQL with pgvector). It provides a unified way to handle document ingestion, vector search, knowledge graphs, and LLM-driven retrieval within a single process model.

How it works

Arcana integrates into the BEAM process model, using Nx.Serving instances for embedders and rerankers under a supervision tree. It supports three modes of operation based on the agentic RAG taxonomy:

  • Advanced RAG: Simple entry points (Arcana.search/2, Arcana.ask/2) that handle query rewriting, hybrid search, and reranking by default.
  • Modular RAG: A composable pipeline (Arcana.Pipeline) where developers can define a custom sequence of steps like rewriting, decomposition, searching, and grounding.
  • Agentic RAG: An LLM-driven loop (Arcana.Loop) where the model chooses tools (search, answer, give up) at runtime to answer complex or multi-hop questions.

Who it’s for

Elixir and Phoenix developers who want to implement RAG capabilities without adding infrastructure overhead or juggling multiple languages (like Python) in their stack.

Highlights

  • Infrastructure-less: Uses your existing Postgres database via pgvector for documents, embeddings, and knowledge graphs.
  • Local-first: Supports local embeddings and cross-encoder reranking via Bumblebee (EXLA, EMLX, Torchx).
  • GraphRAG: Implements entity extraction, community detection (Leiden), and graph search fused with vector results.
  • Linguistic Grounding: Includes tools for NLI hallucination checks and faithfulness scoring.
  • Pluggable Architecture: Every step (chunkers, embedders, searchers, rerankers) is defined as a behaviour, making them easily replaceable.

Related

  • Project
    HKUDS/LightRAG

    A lightweight, graph-based RAG framework that combines knowledge graphs and vector embeddings to provide efficient, high-quality retrieval and global context understanding.

  • Project
    DevYangJC/Argus

    Argus is an open‑source, enterprise‑grade RAG platform that ingests private documents, builds both vector and keyword indexes, and answers questions with a LLM that must cite real sources. It includes a Spring AI‑based ReactAgent for chat, hybrid retrieval with RRF fusion, multi‑level evidence scoring, role‑based security, and a Vue 3 front‑end.

  • Project
    Bessouat40/RAGLight

    A lightweight and modular Python library for implementing Retrieval-Augmented Generation (RAG), allowing users to easily integrate various LLMs and vector stores to chat with their documents.

  • Project
    ChuckHend/pg_vectorize

    A Postgres server and extension that automates text-to-embedding transformation and orchestration for vector, full-text, and hybrid search to enable RAG.

  • Project
    vespa-engine/sample-apps

    A collection of reference applications and tutorials for Vespa, demonstrating how to implement hybrid search, RAG, and multimodal vector retrieval at scale.