Are these terms standardized?

No. Different vendors use them differently. This article describes the distinction Privian uses; expect variation across the ecosystem.

Which one is Privian?

An LLM gateway with a strong focus on prompt privacy. We do not yet handle non-LLM AI workloads like image generation or speech.

Do I need an AI gateway if I only call one model?

Probably not. A gateway becomes valuable when you have multiple callers, multiple providers, or strict requirements around what data leaves your perimeter.

LLM gateway vs. AI gateway

The short version

An LLM gateway is scoped to large language model traffic — chat, completions, embeddings. An AI gateway is broader and typically includes image, audio, and other model modalities. They share the same architectural idea (a chokepoint with policy) but cover different surface areas.

What an LLM gateway is for

Text in, text out
Provider routing (OpenAI, Anthropic, Google, etc.)
Prompt-level policy: masking, validation, allowed models
Observability scoped to text request metadata

What an AI gateway adds

Multimodal handling — images, audio, video
Sometimes a model serving layer of its own
Broader content policy across modalities
Usually a larger surface area and more configuration

Why the distinction matters when you pick one

If 95% of your AI traffic is text chat, an LLM-focused gateway gives you a sharper tool. The policies, the masking detector, the observability dashboards, and the docs are all about text. An AI-gateway abstraction that has to cover image generation and voice as well often makes text harder than it needs to be.

If your roadmap genuinely needs multimodal — image moderation, audio transcription with PII controls, mixed pipelines — the broader category is the right fit.

How Privian fits

Privian is an LLM gateway. The current beta covers text-only workloads with a strong emphasis on prompt privacy: detection, deterministic masking, in-memory rehydration, zero retention of prompt bodies. Multimodal is out of scope today; see Updates for the roadmap.

Frequently asked questions

Are these terms standardized?: No. Different vendors use them differently. This article describes the distinction Privian uses; expect variation across the ecosystem.
Which one is Privian?: An LLM gateway with a strong focus on prompt privacy. We do not yet handle non-LLM AI workloads like image generation or speech.
Do I need an AI gateway if I only call one model?: Probably not. A gateway becomes valuable when you have multiple callers, multiple providers, or strict requirements around what data leaves your perimeter.