core-models

name	core-models
description	Work on core LLM, embeddings, vector store, provider, cache, prompt, and retriever factories in genai-tk. Use when editing genai_tk/core or provider configuration.

GenAI Toolkit Core Models

Read First

• docs/core.md
• docs/llm-selection.md
• genai_tk/core/factories/llm_factory.py
• genai_tk/core/factories/embeddings_factory.py
• genai_tk/core/factories/retriever_factory.py
• config/providers/llm.yaml
• config/providers/embeddings.yaml

Main Components

Component	Code	Tests
LLM creation	genai_tk/core/factories/llm_factory.py	tests/unit_tests/core/test_llm_factory.py
Embeddings	genai_tk/core/factories/embeddings_factory.py	tests/unit_tests/core/test_embeddings_factory.py
Vector stores	genai_tk/core/embeddings_store.py, genai_tk/core/vector_backends/	tests/unit_tests/core/test_embeddings_store.py
Retrievers	genai_tk/core/factories/retriever_factory.py, genai_tk/core/retrievers/	tests/unit_tests/core/test_retriever_factory.py
Cache	genai_tk/core/cache.py	tests/unit_tests/core/test_cache.py
Prompts	genai_tk/core/prompts.py	tests/unit_tests/core/test_prompts.py

Patterns

• Model IDs use model@provider or config aliases such as default and fast_model.
• Keep provider-specific branching inside factories or provider adapters.
• Prefer YAML aliases in config/providers/*.yaml for new models.
• Use fake providers or the pytest profile for unit tests.
• Structured output should use Pydantic models and with_structured_output when available.

Verification

GENAITK_PROFILE=pytest uv run pytest tests/unit_tests/core -q
uv run cli core llm "hello" --llm fast_model

Use real provider tests only when explicitly validating live integration behavior.