The Perfect AI Setup: Unifying Ollama, CLIs, and Apps

To build the ultimate AI development environment, you need three things:

1. A Private Engine (for sensitive data).
2. A Power Engine (for complex reasoning).
3. A Unified Interface (to use them both).

In this guide, we will use switchAILocal to wire up Ollama (Private) and Gemini/Claude (Power) into a single, seamless workstation.

Step 1: The Foundation (switchAILocal)

First, we need the router that will manage traffic between your apps and your models.

git clone https://github.com/traylinx/switchAILocal.git
cd switchAILocal
./ail.sh start

This starts a server at http://localhost:18080 that is ready to accept traffic.

Step 2: The Private Layer (Ollama)

For tasks involving private code or personal data, nothing beats a local model.

1. Install Ollama from ollama.com.
2. Run ollama serve in a terminal.
3. Pull a model: ollama pull llama3.

switchAILocal Auto-Discovery: You don’t need to configure anything. The gateway scans your ports, finds Ollama, and instantly registers ollama:llama3 as an available model.

Check it: curl http://localhost:18080/v1/models -> You’ll see ollama:llama3.

Step 3: The Power Layer (Keyless CLIs)

For heavy lifting, we’ll connect Google and Anthropic models. But instead of generating insecure API keys, we’ll use their secure CLIs.

1. Gemini: Install the Google Cloud SDK and run gcloud auth application-default login.
2. Claude: Install the Anthropic CLI and run anthropic login.

The “Process Wrapper” Magic: switchAILocal detects these binaries in your system path. It automatically creates providers like geminicli:gemini-pro. When you use them, it “wraps” the CLI command, using your existing system-level authentication. Result: High-power cloud AI with zero API keys stored in your gateway config.

Step 4: The Interface (Your IDE)

Now, bring it all together. Open Cursor, VS Code, or Aider.

• Base URL: http://localhost:18080/v1
• API Key: sk-any (Authentication is handled locally).

The Result: A “Hybrid” Workflow

You now have a setup that is the envy of power users everywhere:

• Privacy when you need it: Select ollama:llama3.
• Power when you want it: Select geminicli:gemini-pro.
• No friction: Never change windows. Never copy-paste an API key again.

Get the Tool on GitHub