# Local-only configuration (fully offline, no API keys needed)
#
# Uses Ollama for LLM and sentence-transformers for embeddings.
# Requires: Ollama running locally (https://ollama.com)
#
# Copy to ~/.config/local-deepwiki/config.yaml or use:
# deepwiki config profile save local
---
llm:
provider: ollama
ollama:
model: qwen3-coder:30b # or llama3.1:8b for less VRAM
base_url: http://localhost:11434
embedding:
provider: local # sentence-transformers, runs on CPU/GPU
local:
model: all-MiniLM-L6-v2 # fast, 384-dim, ~80MB
# Embedding cache (avoids re-embedding unchanged files)
embedding_cache:
enabled: true
ttl_seconds: 604800 # 7 days
# LLM response cache (avoids re-generating docs for unchanged code)
llm_cache:
enabled: true
ttl_seconds: 604800
# Wiki generation
wiki:
max_file_docs: 500 # max files to document
max_concurrent_llm_calls: 2 # keep low for local LLM
codemap_enabled: true
codemap_max_topics: 5
# Deep research (local models work best with "quick" preset)
deep_research:
preset: quick
