.third_party/local-llm-research/ for
offline reference. They are not imported into the runtime and are ignored by
Git. The current research set is:
Arka’s local-model selector already uses installed Ollama models and hardware
headroom. The next safe integrations are adapter-level: improve model
discovery, expose llama.cpp/vLLM health probes, and reuse repository-map and
verification patterns. Do not copy source code or vendor dependencies without
reviewing each repository’s license and notices; transitive model/runtime
licenses may differ from the project license.
The checkouts were made as shallow clones for research and should be refreshed
before relying on version-specific behavior.