Privacy-first LLM gateway for AI applications

Privian runs masking and rehydration before any provider call. See the glossary entry for a longer definition, or the AI Security Layer category page for how it fits into the broader stack.

A pure routing gateway changes nothing about what the provider sees. Customer emails, internal documents, support transcripts and developer scratch prompts all reach the model verbatim. A privacy-first gateway shrinks that exposure surface at the edge.

• Customer identifiers in product AI features
• Employee data in internal copilots
• Ticket content in customer support workflows
• Secrets and API keys in developer experimentation

1. Step 1

   Client → Privian

   Your app POSTs prompt + model to a single Privian endpoint with a gateway API key.

2. Step 2

   Detection

   Inbound prompt is scanned for personal and sensitive entities.

3. Step 3

   Masking

   Detected values are replaced with deterministic placeholders (PERSON_1, EMAIL_2, ...) for the request.

4. Step 4

   Provider call

   Masked prompt is forwarded to the configured provider using your decrypted BYOK credentials.

5. Step 5

   Rehydration

   The provider response is scanned and placeholders are restored from the in-memory mapping.

6. Step 6

   Response

   Your app receives a normal, rehydrated response. The mapping is discarded.

Privian is OpenAI Chat Completions-compatible — point the OpenAI SDK (or any HTTP client) at https://api.privian.io/v1/chat/completions.

curl -sS -X POST "https://api.privian.io/v1/chat/completions" \
  -H "Authorization: Bearer $PRIVIAN_GATEWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-4o-mini",
    "messages": [
      { "role": "user", "content": "Email me at michael@example.com about ticket #4821." }
    ]
  }'

Provider-namespaced model ids from the Privian catalog. The legacy /v1/gateway endpoint with the { prompt, model } shape also remains supported. See the API reference for the full request and response shape.

Application sends: "Email michael@example.com about ticket #4821."

Provider receives: "Email EMAIL_1 about ticket #4821."

• Multi-provider routing

  OpenAI, Anthropic and other providers behind one endpoint.

• PII masking

  Personal and secret entities replaced before the provider call.

• Rehydration

  Placeholders restored in the response so your app sees real values.

• BYOK

  Your provider credentials, encrypted at rest, used at request time.

• Gateway API keys

  Scoped Privian keys (sk-gw_live_… / sk-gw_test_…) for your apps.

• Structural observability

  Counters and metadata without storing raw prompts.

• No streaming yet (stream: true is rejected)
• No tool / function calling
• No JSON mode / structured outputs
• No multimodal content (text only)
• No Responses API
• No HIPAA / SOC 2 / PCI claims at this time

Gateways fall into two broad categories. Both are legitimate — they solve different problems, and many teams eventually need both sets of capabilities.

Privian sits in the second category. If your blocker is cost and failover, a routing gateway may be enough; if it is sensitive data reaching model providers, content-level controls are the part that matters.

A short framework for comparing gateways during a technical evaluation. Test each dimension against your own traffic rather than vendor descriptions.

Privian's answers to these are published on the data path, supported models and Trust Center pages, with side-by-side framing in the comparisons.

What is an LLM gateway?

An LLM gateway is a routing layer that accepts AI requests from your application and forwards them to one or more model providers. It is where authentication, model selection, routing and observability live. Privian adds PII detection, masking and rehydration on top of that routing layer.

Why use an LLM gateway?

A gateway gives you one integration point instead of one per provider: shared authentication, provider-agnostic model ids, consistent observability, and a single place to enforce data controls. Without it, every provider change and every privacy control has to be re-implemented in application code.

Can an LLM gateway mask PII?

Only if the gateway implements it. Routing-only gateways forward whatever your application sends. Privian detects supported sensitive entities in the inbound prompt, replaces them with deterministic placeholders before the provider call, and rehydrates the response in memory.

What is the difference between an LLM gateway and an AI gateway?

The terms overlap. "AI gateway" is usually the broader term for any proxy in front of AI providers, often focused on routing, caching, rate limiting and cost control. "LLM gateway" specifically refers to the layer in front of large language model providers. Privian is an LLM gateway with privacy and data-control capabilities in the request path.

Can Privian route requests to multiple LLM providers?

Yes. Models are addressed as provider/id — for example openai/gpt-4o-mini or anthropic/claude-sonnet-4-5 — so one endpoint and one client integration can reach several providers. See the supported models page for the current catalog.

Does Privian protect prompts before they reach model providers?

Yes. Detection and masking run at the gateway before the outbound provider call, so only masked content leaves the boundary for entities the detector supports. Raw prompts and responses are not persisted.

Can I use Privian with OpenAI and Anthropic?

Yes, using your own provider credentials (BYOK). Keys are stored AES-GCM encrypted at rest and decrypted in-process at request time; provider token usage continues to bill against your own provider account.

Is an LLM gateway useful for AI SaaS companies?

It is often the fastest way to answer enterprise security questions about AI data flows: one documented boundary where prompts are inspected and protected, provider credentials are held, and retention behaviour is defined — rather than controls scattered across application code.

How is Privian different from a generic AI gateway?

A generic AI gateway routes traffic — whatever you POST, the provider sees. Privian is privacy-first: sensitive values are detected and replaced with deterministic placeholders before the provider call, then restored in the response.

Does Privian support the OpenAI SDK as a drop-in?

Yes — for non-streaming Chat Completions. Point the OpenAI Python or Node SDK at https://api.privian.io/v1 as the base URL and call /chat/completions with model + messages[]. Streaming, tool/function calling, JSON mode, multimodal content, and the Responses API are not yet supported.

How does BYOK work?

You add your provider credentials (OpenAI, Anthropic, etc.) in the dashboard. They are encrypted at rest using AES-GCM and decrypted in-process at request time. Token usage continues to bill against your provider account — Privian does not resell tokens.

Does the gateway store prompts or responses?

No. Raw prompts and responses are never persisted. Only structural counters are kept for observability.

• Category: AI Security Layer
• Category: Prompt Privacy
• Glossary: What is an LLM gateway?
• Concepts: BYOK
• Authentication
• Quickstart
• Architecture overview
• API reference
• PII masking
• Data path (per-hop)
• Compare: Privian vs Portkey
• Compare: Privian vs LiteLLM
• Compare: Privian vs Cloudflare AI Gateway
• Compare: Privian vs Kong AI Gateway
• Compare: Privian vs Skyflow
• Trust Center
• Privian Blueprint (evaluation package)
• Pricing and packaging
• Enterprise AI procurement framework
• Enterprise-Ready AI for startups
• AI vendor due diligence checklist
• Buyer's guide: how to choose an LLM gateway
• Migrate existing LLM traffic to Privian
• Solution: OpenAI data privacy
• Solution: Claude data privacy