Knowledge Basics

Odeus processes knowledge in different ways. Based on file size and context, the system automatically chooses between two main methods: preview processing in the context window for smaller documents and embedding search for large files.

File Preview in the Model's Context Window

Users get the best results when the model can process the entire text of the knowledge. That's why as much text as possible is sent to the model as a preview. For small files, this can be the entire document.

Embedding Search

Additionally, the model can use an embedding search. During upload, the file is first split into smaller sections (chunks), which are then converted into embeddings. Each embedding receives a vector — a numerical sequence that describes various thematic dimensions.

When a question is asked, the system performs a vector search to identify the relevant vectors and their underlying sections. This doesn't search for specific words (like a keyword search with Ctrl + F), but for sections with semantic similarity to the query.

Example: If the embedding search is looking for "bread," it will also find sections about "baguette," even if the word "bread" doesn't appear.

Only these relevant parts are sent to the model in context. This enables working with very large documents that exceed the model's context window.

Retrieval pipeline

Odeus's retrieval is configurable by admins, so it can be matched to your content and language needs:

• Embedding Model — Admin-configurable (e.g. OpenAI 3-small/large, Jina, Cohere, Voyage), including a self-hosted OpenAI-compatible endpoint.

• Chunking — Multi-resolution chunking, with entity extraction and image indexing for richer retrieval.

• Reranking — An optional reranker (Cohere, Jina, or Voyage) reorders results for relevance before they reach the model.

Retrieved passages come back with inline citations to the source file, page, and location.