Infrastructure

Settings → Infrastructure is where you repoint GCTRL at external storage and, most importantly, at GPU-accelerated inference. This is the single highest-impact configuration page for performance.

Pointing at external services

GCTRL ships with bundled Neo4j, Qdrant, Postgres, and Ollama, but you can point it at your own. Every endpoint is env-overridable and configurable from Settings → Infrastructure:

HTTP services (Qdrant, Ollama) re-target on the fly. Pooled database connections (Neo4j, Postgres) hold a connection pool, so changes take effect after a restart.

The big one: switch to native Ollama for GPU acceleration

The bundled Ollama runs inside Docker. On macOS / Apple Silicon, Docker cannot access the Apple GPU (Metal) - so the containerized Ollama runs CPU-only, leaving the machine's GPU and unified memory completely unused. On a capable Mac this is the difference between sluggish and fast.

The fix: run Ollama natively on the host and point GCTRL at it. One setting repoints both the RAG/agent path and KEX extraction/embeddings to the GPU-accelerated native Ollama.

macOS / Apple Silicon

1. Install Ollama natively:

   brew install ollama
   

   (or install the app from ollama.com).

2. Pull a model:

   ollama pull qwen2.5:7b
   

3. Repoint GCTRL. In Settings → Infrastructure, set the Ollama base URL to:

   http://host.docker.internal:11434
   

   host.docker.internal lets the containers reach the host's native Ollama. The change applies immediately (Ollama is an HTTP service).

That one switch routes both:

• the RAG / agent path (Talk-to-Graph generation), and
• KEX extraction and embeddings

to the native, GPU-accelerated Ollama - recovering the Apple GPU and unified memory the Docker path could never touch.

Linux

The same idea applies on Linux: run a host-native Ollama with a GPU (e.g. an NVIDIA card) instead of the bundled container, then set the Ollama base URL in Settings → Infrastructure to your host-native Ollama endpoint. As on macOS, the one switch repoints both the RAG/agent path and KEX extraction/embeddings.

See also

• Performance - the full best-performance guide, where native Ollama is the biggest lever
• LLM Providers - model and provider selection