This guide provides step-by-step instructions on how to bridge your local Ollama instance with Crowdin.

Since Crowdin is a cloud-based platform, it cannot directly access your localhost. To bridge this gap, you can use ngrok to create a secure gateway.

Step 1: Prepare Ollama

First, ensure Ollama is running and configured correctly on your local machine.

1. Start Ollama. Ensure the application is active in your system tray or terminal.

2. Download a Model. Use a model that performs well with translations. Command example:

ollama pull llama3

3. Verify local access. Open http://localhost:11434 in your browser. You should see the message: "Ollama is running".

Step 2: Create a Tunnel with ngrok

This step creates a temporary public URL that Crowdin can use to "talk" to your computer.

1. Install ngrok. Download it from ngrok.com.

2. Start the tunnel. Run the following command in your terminal:

ngrok http 11434

3. Copy the Forwarding URL. Look for the line starting with Forwarding. It will look like https://a1b2-c3d4.ngrok-free.app.

Note: Keep this terminal window open. If you close it, the connection will break.

Step 3: Connect to Crowdin Store

Now, we will use Crowdin's native OpenAI connector to point toward your local machine.

1. Open Crowdin. Navigate to your project and go to Settings > AI.

2. Select Provider. Choose the OpenAI (Native) connector.

3. Configure Connection:

• API Key: Enter any placeholder text (e.g., ollama). While Ollama doesn't require a real key, Crowdin's form requires this field to be filled.
• Base URL: Paste your ngrok URL and add /v1 at the end.
• Correct format: https://your-unique-id.ngrok-free.app/v1
• Model Name: Type the exact name of the model you downloaded.

Note

• On the free version of ngrok, your URL changes every time you restart the app. You will need to update the Base URL in Crowdin whenever you restart your tunnel.
• Local LLMs depend on your hardware. If the model takes too long to respond, Crowdin might report a timeout error.
• Using this method keeps your data processing on your local hardware, which is excellent for sensitive content.