Siper Flow is an LLM proxy that masks personal data — including Turkish IDs and addresses — at runtime, then cuts token costs with exact and semantic cache. Works with OpenAI, Azure, Anthropic, Gemini and Groq; only the base URL changes.
NER catches context-dependent data such as addresses; regex catches fixed-format data such as TCKN. Repeat questions are answered from the semantic cache.
Personal data is masked before it reaches any provider; the token savings come on top of that.
NER and regex detect personal data in every request and mask it before it reaches the provider — Turkish IDs, card numbers and names included.
Exact cache for identical requests and semantic cache for similar ones, including Turkish questions. Repeat calls are answered in milliseconds without reaching the provider.
Removes redundant tokens from user messages while leaving the system prompt untouched, so the same answer costs fewer tokens.
Real-time view of cache hits, masked fields and token savings per tenant.
Bring your own provider keys and run Flow inside your own infrastructure, so your data stays within your boundary.
One endpoint for OpenAI, Azure, Anthropic, Gemini and Groq. Only the base URL changes in your application.
