ra-h-os/OLLAMA-LOCAL-PROFILE.md

Ollama Local Profile

Ollama is the convenience runtime path for the supported local profile.

Start Ollama and pull the supported model pair:

ollama serve
ollama pull qwen3:4b
ollama pull qwen3-embedding:0.6b

Configure RA-H:

LLM_PROFILE=openai-compatible
LLM_BASE_URL=http://127.0.0.1:11434/v1
LLM_MODEL=qwen3:4b

EMBEDDING_PROFILE=openai-compatible
EMBEDDING_BASE_URL=http://127.0.0.1:11434/v1
EMBEDDING_MODEL=qwen3-embedding:0.6b
EMBEDDING_DIMENSIONS=1024

Validate:

npm run doctor:local-ai

Changing EMBEDDING_MODEL, EMBEDDING_DIMENSIONS, EMBEDDING_PROFILE, or VECTOR_BACKEND requires a vector rebuild:

npm run rebuild:embeddings

Local utility LLM quality can affect descriptions, extraction summaries, transcript summaries, and edge inference. Keep custom model overrides experimental until they pass your own workflow checks.