openwikiView on GitHub
A beige vintage computer terminal with a soft green glow and a brass desk key.

Any LLM

Gemini, OpenAI-compatible, or fully offline

The LLM reads each source and writes the summary, claims and entity and topic lists. Pick whichever fits your budget and privacy needs; put the settings in .env in your workspace.

Google Gemini

GEMINI_API_KEY=...
# optional, this is the default:
GEMINI_MODEL=gemini-3.1-flash-lite

OpenAI, OpenRouter, Groq

LLM_PROVIDER=openai
OPENAI_API_KEY=...
OPENAI_MODEL=<model id>
OPENAI_BASE_URL=https://openrouter.ai/api/v1   # or Groq, or leave unset for OpenAI

Ollama or LM Studio, fully offline

LLM_PROVIDER=openai
OPENAI_BASE_URL=http://localhost:11434/v1   # LM Studio: :1234/v1
OPENAI_MODEL=llama3.1

No key is needed for a local server, and nothing leaves your machine except the caption and article downloads. Larger models extract entities and topics more reliably.

No model at all

Pass --analysis-file with precomputed JSON and OpenWiki builds the pages without calling any LLM.