Retrieval-Augmented Generation (RAG) Database

Soliplex depends on the haiku-rag library to manage its retrieval-augmented generation (RAG) searches. That library stores its extracted documents / chunks / embeddings in LanceDB databases.

The example installation of Soliplex uses the Soliplex documentation as its RAG corpus, and expects that database to be created at db/rag/rag.lancedb.

Note on haiku-rag Versions

The soliplex code itself requires only the haiku-rag-slim project (https://pypi.org/project/haiku.rag-slim/), which allows for queries aganst an existing LanceDB database.

However, this dependency is not sufficient to perform the ingestion / indexing of documents. For that purpose, either:

• Install the main haiku-rag project https://pypi.org/project/haiku.rag/ wihch will pull in all the dependencies required to ingest and index documents.

• Pull the docling-serve Docker image, and run its server, with your haiku.rag.yaml file configured to use it.

See the haiku.rag documentation to determine:

• Which installation do you need?

• What are the tradeoffs of local vs. remote processing?

• How to configure haiku-rag to run in "remote processing" mode?

Adding a single document

export OLLAMA_BASE_URL=<your Ollama server / port>
haiku-rag --config example/haiku.rag.yaml \
  add-src --db db/rag/rag.lancedb docs/index.md
...
Document <UUID> added successfully.

Adding all documents in a directory

export OLLAMA_BASE_URL=<your Ollama server / port>
haiku-rag --config example/haiku.rag.yaml \
  add-src --db db/rag/rag.lancedb docs/
...
17 documents added successfully.

Configuration of haiku-rag clients within Soliplex

Please see this page for notes on configuring the various haiku-rag clients used in a Soliplex installation.