Setting up Providers

Karpa supports multiple AI backends. Local runtimes (LM Studio, Ollama) keep everything on your machine; cloud providers trade some privacy for convenience. Here is how to set up each one.

LM Studio (Local)

LM Studio is the recommended local runtime for translation work.

1. Install LM Studio

Download from lmstudio.ai and install for your platform (macOS, Windows, Linux).

2. Download a model

1. Open the Search tab in LM Studio.
2. Search for a translation-capable model. Good starting points include HY-MT1.5-7B or a small instruct model such as llama3.2 (see Choosing a Model).
3. Download a quantized variant such as Q4_K_M.

Quantized models are smaller and faster while keeping good translation quality. Q4_K_M is the usual sweet spot for local use.

3. Start the local server

1. Go to the Developer / Local Server tab.
2. Load your downloaded model.
3. Set the port to 1234.
4. Click Start Server and confirm the status shows running.

4. Connect Karpa

In Settings → Connection, select LM Studio. The default URL (http://localhost:1234) works if you kept the default port. Use the model picker to select your loaded model, then click Test Connection.

Ollama (Local)

Ollama is an alternative local runtime.

1. Install Ollama

Download from ollama.com and install.

2. Pull a model

ollama pull llama3.2
# or any other model that works well for translation

3. Make sure the server is running

Ollama starts a background server automatically on port 11434 when you use the CLI. Verify with:

curl http://localhost:11434/api/tags

4. Connect Karpa

In Settings → Connection, select Ollama. The default URL (http://localhost:11434) works out of the box.

OpenAI (Cloud)

1. Create an account at platform.openai.com.
2. Generate an API key in the dashboard.
3. In Karpa's Settings → Connection, select OpenAI.
4. Enter your API key and choose a model (default: gpt-4o-mini).

Anthropic (Cloud)

1. Create an account at console.anthropic.com.
2. Generate an API key.
3. In Karpa's Settings → Connection, select Anthropic.
4. Enter your API key and choose a model (default: claude-sonnet-4-20250514).

Google Gemini (Cloud)

1. Get an API key from aistudio.google.com.
2. In Karpa's Settings → Connection, select Google Gemini.
3. Enter your API key and choose a model (default: gemini-2.0-flash).

OpenRouter (Cloud)

One API key, 200+ models from many providers.

1. Create an account at openrouter.ai.
2. Generate an API key.
3. In Karpa's Settings → Connection, select OpenRouter.
4. Enter your API key and pick any model from the catalog.

Custom (Self-hosted / Proxy)

Any OpenAI-compatible endpoint works: vLLM, LiteLLM, text-generation-webui, corporate gateways, you name it.

1. In Settings → Connection, select Custom.
2. Enter the base URL of the endpoint.
3. Add an API key if your endpoint requires one.
4. Type the model identifier manually if auto-fetch cannot list models.

That's it. You can switch providers anytime from Settings without restarting the app.