Skip to main content

Lucenia

The Search AI Engine for Private Clouds

Search, retrieval and AI over your own data — extraction through grounding, in one engine, on your own infrastructure.

From nothing to a search result

One dependency and no cluster to provision. A new node runs for fourteen days without a licence, so you can reach a result before deciding whether to go further.

Take the quickstart →
# Run it
export LUCENIA_INITIAL_ADMIN_PASSWORD="<the password you set in docker-compose.yml>"
docker compose up -d

# Check it
curl -k -u admin:$LUCENIA_INITIAL_ADMIN_PASSWORD "https://localhost:9200/"

Start with what you are trying to do

Search by meaning

Send plain text and let the cluster find the closest matches — no client-side embedding calls, no second service to keep in step.

semantic query

Process your content

Extract, chunk, embed and OCR documents in more than ten formats — PDF, DOCX, HTML, GeoTIFF — on the way in.

ingest pipelines

Work with places

Shapes, not just points. Route by road rather than radius, summarise where data concentrates, and render vector tiles from the cluster.

geospatial intelligence

Give agents memory

A durable remember and recall layer with recency-aware retrieval, per-organisation isolation, and drift anchors.

inference memory

Govern sensitive data

Detect and redact PII, PHI, payment and secret data before it is indexed, against named policies you define.

compliance engine

Run it in production

Docker, Kubernetes with Helm, tarball or Windows — with TLS, attribute-based access control and audit logging.

install and configure

Why teams choose it

It runs where your data is

Every component runs on your infrastructure. Content is extracted, embedded, redacted and searched inside your own trust boundary — including air-gapped deployments.

One engine, not a pipeline of services

Extraction, chunking, embedding, vector and keyword search, reranking and grounding are the same system. There is no separate service to deploy, scale and keep in step with your mappings.

Answers you can defend

Retrieve by a vocabulary you control rather than by an embedding you do not, and carry provenance through to the citation. "The ontology says so" is a reason; "the vector said so" is not.

Looking for something specific?

Every field type, query, aggregation, processor and API is in the reference.