Which models work with Vanta?

Vanta supports OpenAI-compatible servers, downloaded GGUF and MLX models, and Apple Intelligence on eligible devices. A supported connection or file format does not guarantee that every model will load, support tools or run comfortably on every device.

Capabilities by runtime

This table describes implemented capabilities. It is not a performance benchmark or a list of configurations tested on physical devices.

Runtime	Where the model runs	Text chat	Images	Model tool calls
OpenAI-compatible server or cloud API	Your chosen server or provider	Supported with a compatible chat API	Requires a vision model and compatible API	Requires support from the model and server
GGUF / llama.cpp	Your iPhone or iPad	Supported model architectures	Not supported in Vanta	Not supported in Vanta
MLX	Your iPhone or iPad	Supported MLX models	Selected models marked as vision-capable	Selected models marked as tool-capable
Apple Intelligence	Apple’s on-device system runtime	When the system model is available	Not supported in Vanta	Supported within system and schema limits

PDF questions use extracted text and retrieval. They do not require image input, although document quality and context size affect the answer. Read the files guide.

iPhone and iPad requirements

The current App Store listing requires iOS 18 or iPadOS 18 or later. That is the app installation floor, not a promise that a particular local model fits.

Apple Intelligence requires iOS or iPadOS 26 or later for Vanta’s system-model integration, compatible hardware, Apple Intelligence enabled, and a supported language with the model ready. The app displays the system’s current availability state.

For downloaded models, available memory matters more than free storage alone. Larger contexts and image inputs increase memory use. Start with a small supported model and a short context, then increase only when the device remains stable.

Self-hosted connections

LM Studio, Ollama and vLLM expose APIs Vanta can use. The server must be reachable from your phone, supply the correct model identifier, and support chat completions in the expected format.

Server	Typical base URL on your network	Guide
LM Studio	http://192.168.1.10:1234/v1	LM Studio setup
Ollama	http://192.168.1.10:11434/v1	Ollama setup
vLLM	http://192.168.1.10:8000/v1	vLLM setup

These are example LAN addresses; use your server’s actual address and port. Custom proxy paths may differ. An API advertised as OpenAI-compatible can still differ in streaming, tool calls, vision or supported parameters.

What this page can tell you

It can tell you which runtimes and capability gates Vanta implements. It cannot predict tokens per second, memory headroom or sustained stability for your exact device and model. No new device benchmarks were collected for this documentation.

If a configuration fails, include the exact device, OS, Vanta version, model, format, quantization and context setting when you contact support.