README.md

Embeddorium

Build retrieval pipelines. Inspect every stage. Compare what works.

Embeddorium is an open-source, local-first retrieval engineering workbench for building, inspecting, and comparing RAG retrieval pipelines.

• Declarative: Define how your data should be ingested, processed, indexed, and searched. Embeddorium executes the pipeline, preserves every intermediate artifact, and makes each processing step available for inspection. Explicit configurations make retrieval workflows easier to reproduce, understand, compare, and debug.

• Configuration-First: Datasets, providers, ingestion pipelines, chunking strategies, embedding models, search methods, fusion, and reranking are all configured independently. Run the same data through different configurations without rewriting application code or rebuilding the entire retrieval stack.

• Plugin-Based: Extend Embeddorium with custom parsers, filters, chunkers, embedding providers, rerankers, and other pipeline components. When a data format, model provider, or processing strategy is not supported, add a plugin, refresh the UI, and use it as part of the same pipeline. Spend less time on integration boilerplate and more time improving retrieval.

• Evaluation-Driven: Build evaluation sets from real queries, mark expected relevant passages, and compare retrieval configurations using metrics such as Recall@K, MRR, and nDCG. Measure individual search cases or aggregate results across a dataset to understand which pipeline configuration performs best.

Quick start

1. Install Docker Compose v2.

2. Prepare the environment:

git clone https://github.com/PerminovEugene/Embeddorium.git
cd Embeddorium
cp .env.docker .env

3. Start the containers:

docker compose up -d

4. Run migration:

docker compose run --rm migrate

4. Open the UI at http://localhost:5173/.

The Quick start guide walks through a complete ingestion and retrieval run using the built-in mock provider, so no model server is required.

Documentation

• Quick start [ready]
• Providers [ready]
• Datasets [ready]
• Ingestion pipelines [todo]
• Retrieval [todo]
• Evaluation [ready]
• Plugins [todo]
• Architecture [todo]
• Configuration [todo]
• API [todo]
• Scope and limitations [todo]
• Supported capabilities [todo]
• Contributing [todo]

Browse the full documentation in docs/index.md [todo].

License

Apache License 2.0. See LICENSE.md.