Local Model Servers Integration Guide

Overview

Local model servers such as Ollama, LM Studio, llama.cpp-compatible servers, and OpenAI-compatible local endpoints are common in privacy-sensitive teams. Cup’n’String discovers local ports, attributes access, and governs which agents may connect.

Support level

Auto-Discovered + Compatibility Adapter

What Cup’n’String detects

• Known local model server processes
• OpenAI-compatible local endpoints
• Common ports and socket patterns

What it governs

• Which IDEs/agents may connect
• Exposure of local endpoints to other processes
• Tunnels and remote access
• Secrets passed to local prompts/tools

Recommended policies

• Restrict local model access to approved agents
• Block remote exposure unless explicitly allowed
• Audit prompt/tool metadata where proxying is enabled

Setup outline

1. Ensure the Cup’n’String agent is active.
2. Fire up your local model server of choice (e.g. Ollama).
3. The compatibility adapter auto-detects the running port (e.g. 11434) and restricts direct socket access from unauthorized processes.

Verification

Try connecting to Ollama from a non-approved terminal agent, and verify that the connection is blocked.

Troubleshooting

If server is not detected, check if it runs on custom non-standard ports, and configure them explicitly in Cup’n’String client settings.

Known limitations

Model inference content visibility depends on routing through managed proxy.