v1.0.2

What's new

Adds an openai-compatible provider so pi-knowledge-search can point at any OpenAI-compatible /v1/embeddings endpoint — llama.cpp, vLLM, litellm, LM Studio, Ollama API mode, etc.

Added

• New openai-compatible provider type with configurable baseUrl (#3)
• KNOWLEDGE_SEARCH_COMPAT_BASE_URL, KNOWLEDGE_SEARCH_COMPAT_API_KEY, KNOWLEDGE_SEARCH_COMPAT_MODEL env vars
• Optional apiKey on openai-compatible (skipped when unset — for runners that don't require auth)

Fixed

• type: "openai" with a custom baseUrl used to be silently ignored and hit api.openai.com — now throws a clear error steering users to openai-compatible (closes #4)
• Env-var-only config path now correctly defaults to the openai provider when OPENAI_API_KEY is set (previously returned provider: null)

Security

• openai-compatible does not fall back to OPENAI_API_KEY — prevents silently leaking a real OpenAI key to a third-party endpoint. Set KNOWLEDGE_SEARCH_COMPAT_API_KEY explicitly.

Thanks @loopyd for the PR and bug report.