Get started with Vanta

You need one working model to start. It can run on your iPhone or iPad, on a computer you control, or at a cloud provider. Vanta supplies the chat interface; you choose the model and where it runs.

1. Choose a starting point

You already have…	Start with…
LM Studio on a computer	The LM Studio guide
Ollama on a computer or server	The Ollama guide
A vLLM server	The vLLM guide
A cloud API key	Cloud providers
No server or API account	On-device models

Apple Intelligence can be a quick starting point on an eligible device with the system model available. A supported downloaded model is another option; its size must fit your device’s memory.

2. Add the source

For a server or custom cloud endpoint, open Settings → Providers → Add provider. Enter the API base address and a key if the server requires one. A local address such as 192.168.1.10:1234 belongs to your computer; localhost refers to the phone itself.

For downloaded models, open Local AI in Settings and choose a supported model or import a GGUF file. Finish its download before starting a local chat.

3. Save and select a model

A preset remembers the model and its settings. Create a preset for your connected provider or downloaded model, then select it in Settings → Models. The model picker in the chat also lets you switch between saved presets.

Use the exact model identifier supplied by a remote server. A friendly preset name is just for you; it does not change the model the server runs.

4. Send a simple first message

Start a new chat and ask: “Say hello in one sentence.” Check that the response finishes. Keep this first check free of files and tools so a connection problem is easy to identify.

After basic chat works, add a PDF, configure MCP tools, or adjust your preset. Choose a vision-capable model before attaching an image.

Before using sensitive information

Read privacy and data flow. A remote model receives the context assembled for the request. A local model keeps inference on-device, but separately configured tools, embeddings and cloud speech can still make network requests.

If the first message fails, use troubleshooting or contact support.