Models and hardware¶
Business Index uses two local Ollama models:
- an embedding model converts document passages and questions into vectors for semantic retrieval;
- a chat model proposes evidence-backed business concepts and prepares an answer from retrieved evidence.
Profiles¶
| Profile | Embedding model | Intended use |
|---|---|---|
| Minimal footprint | all-minilm |
Smaller computers and the lowest storage use |
| Lightweight | nomic-embed-text |
Laptops where responsiveness matters most |
| Balanced | nomic-embed-text |
Recommended starting point for most users |
| Higher quality | nomic-embed-text |
Computers with more memory available to Docker |
The selected profile also chooses an appropriate chat model. Exact model sizes can change between Ollama releases, so Business Index shows locally installed models rather than promising a fixed download size.
Practical guidance¶
- Start with Balanced on a computer with 16 GB or more memory.
- Choose Minimal footprint or Lightweight when Docker has limited memory.
- A larger model may improve wording and evidence use, but it does not make unsupported facts available.
- Model download and first indexing can take time on a slower connection or CPU.
Docker Desktop may have less memory available than the computer's total memory. If model preparation fails, review Docker Desktop's resource settings or choose a smaller profile.
Changing the embedding model rebuilds derived semantic representations. Changing the chat model re-runs local concept and relationship extraction. Neither change discards durable business-concept identity or feedback history.