Description

Ollama is detected because localhost:11434 responds, but onboarding later fails because containers cannot reach host.openshell.internal:11434.

Reproduction Steps

1. Start from an Ubuntu machine with Docker CE installed and running.
2. Install Ollama.
3. Start Ollama in a host-only way so it binds to loopback, for example using the default system service or any mode that results in:
   ss -ltnp | grep 11434
   showing:
   127.0.0.1:11434
4. Verify that Ollama appears healthy from the host:
   curl http://localhost:11434/api/tags
5. Ensure at least one Ollama model exists, for example:
   ollama pull nemotron-3-nano:30b
   ollama list
6. Run the NemoClaw installer:
   curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
7. During onboarding:
   enter a valid sandbox name
   choose Local Ollama
   choose the available Ollama model
8. Continue until onboarding reaches inference provider setup.

Environment

• Openshell: 0.0.11
• nemoclaw: 0.1.0 (main: dbfd78c)
• OS: Ubuntu
• Container runtime: Docker CE
• GPU: Available
• Ollama: Installed
• Ollama models: None (ollama list is empty)
• NemoClaw install method: curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash

Debug Output

Logs

Checklist

• I confirmed this bug is reproducible
• I searched existing issues and this is not a duplicate

[NVB#6000934]