Local embeddings
bge-m3 computes embeddings on your CPU. This indexing step does not send document text to an external AI provider.
Applications → Ollama
Local indexing, included in Office.
Office includes local document indexing and search: embeddings are computed on your server, without a paid AI option. For AI-written answers, connect a model using a compatible API key or the Managed AI option, subject to availability.
What Ollama does
Ollama runs bge-m3 on your Office server to turn document text into searchable vectors. Embedding a document does not write an answer.
Why Ollama on ReefOffice
Search by meaning
Qdrant stores your index locally and helps find relevant passages, even when the wording differs.
Written answers: a separate step
A compatible API key or Managed AI, subject to availability, supplies generation. Your selected provider receives the prompt and context sent for that answer.
Multilingual search
bge-m3 supports multilingual retrieval. Quality depends on documents, extraction and query; always check the source.
Indexing without a paid AI option
Office includes the local embedding runtime, not a generative chat model or dedicated GPU.
Connected to your documents
Ollama supplies embeddings for Open WebUI and document-indexing recipes. Hermes uses a separately configured generation route.
Technical detail
How it works
Ollama runs as a native NixOS service. bge-m3 computes embeddings; Qdrant stores the local index. Initial loading and indexing use CPU, RAM and disk.
How it fits your stack
Open WebUI retrieves passages through local embeddings; writing answers uses the separately configured generation route. Ollama is internal, not a public endpoint.
Frequently asked questions
How long does indexing take? +
Duration depends on document volume, extraction and resources. The first model load can take longer. No fixed duration or instantaneous answer is promised.
How much storage do AI models use? +
The embedding model, extracted text and vectors use server storage. Capacity depends on volume and configuration; this Office inclusion does not preinstall a generative chat model.
How do I get AI-written answers? +
Connect a compatible API key or Managed AI when available. Prompts and selected context reach the chosen provider under its terms. Local indexing remains available without this option.
Explore local document search
Discover the included index and its sources in a demo. AI-written answers depend on the generation route available for that session.