Same anonymization engine, three deployment shapes. Pick based on how much control you want over outbound payloads and where the proxy lives.

Mode 1 — Transparent Proxy

One endpoint change. We handle everything.

You point your OpenAI SDK at https://gate.stellarbase.ai/v1. From your app’s perspective, nothing else changes — it looks like OpenAI, Anthropic, or Google. Behind the scenes: StellarGate anonymizes the prompt, forwards to the provider, resolves tokens on the response, returns. Everything in one round-trip.

Flow

1. Your app sends a chat completion request to StellarGate
2. StellarGate scans the prompt, replaces sensitive entities with reversible tokens
3. Sanitized prompt forwarded to the LLM provider (OpenAI, Anthropic, etc.)
4. LLM responds with tokens in place of real data
5. StellarGate substitutes tokens with original values
6. Final response returned to your app

When to pick Mode 1

• You want a drop-in replacement for your existing LLM client
• You don’t need to inspect the sanitized prompt before it’s sent
• You’re fine with us running the proxy in our EU cloud
• You want the simplest possible integration — literally one URL change

Pricing

€0.10 / 1M tokens on the anonymization engine + LLM provider cost passed through at their public rate (no markup from us).

Mode 2 — Tokenized Handoff

We anonymize. You call the LLM. We de-anonymize.

Same anonymization engine, but split into two API calls with your app in the middle. You get full visibility into every outbound payload — useful when compliance requires you to log or inspect exactly what’s sent to external LLMs.

Flow

1. Phase 1 (anonymize): your app sends the prompt to StellarGate. StellarGate returns the sanitized prompt + a map_id.
2. Phase 2 (call LLM): your app sends the sanitized prompt to the LLM provider directly. Log it, inspect it, transform it — you’re in control.
3. Phase 3 (de-anonymize): your app sends the LLM’s response + the map_id back to StellarGate. StellarGate returns the response with tokens resolved.

When to pick Mode 2

• Your compliance team requires you to log every outbound payload
• You want to transform or filter the sanitized prompt before sending
• You use LLM providers we don’t have a direct integration with
• You want to A/B test different providers on the same anonymized input

Pricing

€0.10 / 1M tokens on the anonymization engine. You pay your LLM provider directly for the completion.

Mode 3 — Self-Hosted

Same as Mode 1, but in your own infrastructure.

Deploy StellarGate as Docker / Helm inside your own network. The proxy runs where your data already lives. No internet egress required for the anonymization itself; the LLM provider call goes out to the public internet from your egress (or from a locally-hosted LLM, for full air-gap).

When to pick Mode 3

• Regulated workloads where even the proxy can’t run on external cloud
• Strict data locality — data cannot cross a network boundary, ever
• Air-gapped deployments (paired with an internally-hosted LLM)
• Very high volumes where per-token billing becomes impractical

Pricing

Annual licence, volume-tiered. Unlimited requests. Contact sales for a quote.

Mode comparison

Property	Mode 1	Mode 2	Mode 3
API calls per interaction	1	3	1
Where StellarGate runs	Our EU cloud	Our EU cloud	Your infrastructure
Outbound payload visibility	Metadata only	Full	Full (inside your DC)
LLM cost	Passed through at cost	You pay provider directly	Your choice / self-hosted LLM
Integration effort	One URL change	SDK, 3 calls	Deployment + SDK
Pricing	Per-token	Per-token	Annual licence

Can I switch modes later?

Yes. Dictionaries, rule sets, HITL policies, and audit logs are all portable across modes. A typical progression: start with Mode 1 for the pilot → move to Mode 2 when compliance wants outbound visibility → graduate to Mode 3 when workloads grow or regulations tighten.

Related

• What We Detect — the entities anonymized in every mode
• Integration — code patterns for each mode
• Air-gapped Deployment — Mode 3 in practice