v1.4.1 — Rename mkg_client to local_llm_client

v1.4.1 — Rename mkg_client.py to local_llm_client.py

Cosmetic rename with a clear docstring. Zero behavioural changes.

Why

The file daemon/mkg_client.py was named after the author's private LLM gateway ("MKG" = Mecha King Ghidorah, a personal MCP router). The name implied a dependency that doesn't actually exist — the module is in fact a plain OpenAI-compatible HTTP client that works with anything speaking that API.

What changed

• daemon/mkg_client.py → daemon/local_llm_client.py (via git mv, history preserved)
• Class MKGClient → LocalLLMClient
• Backward-compat alias retained — from daemon.mkg_client import MKGClient no longer works since the file moved, but from daemon.local_llm_client import MKGClient still resolves to the new class. Safe to remove in a future major version.
• New module docstring explains clearly:
  • What this is: optional helper for users who want LLM-generated summaries stored alongside ChromaDB embeddings during indexing.
  • When it's used: only when ConversationIndexer(enable_mkg_analysis=True) — defaults to False everywhere in-tree.
  • What endpoints work: llama.cpp, vLLM, LM Studio, Ollama (via OpenAI proxy), NVIDIA NIM, Groq, OpenAI itself — anything OpenAI-compatible.
  • How to configure: AGENT_GENESIS_LLM_ENDPOINT env var.
  • Why the kwarg name is unchanged: enable_mkg_analysis stays for backward compatibility with all downstream callers.

No behavioural changes

• Default endpoint, timeout, HTTP behaviour all identical to v1.4.0.
• All existing imports that used the backward-compat path (e.g. from daemon.mkg_client import MKGClient) need to be updated to from daemon.local_llm_client import MKGClient — but the class name alias means no logic changes are needed in calling code.
• The public ConversationIndexer API is unchanged. enable_mkg_analysis kwarg and the /index/trigger JSON enable_mkg key both still work.

Upgrade

Drop-in for anyone on v1.4.0. If you were directly importing from daemon.mkg_client, update to daemon.local_llm_client. Everyone else: no action needed.