ThriceGate

LLM gateway: one API for DeepSeek, Sansi, and a prompt guard

An LLM gateway is the layer between your code and the models. Ours is OpenAI-compatible and ships two things most gateways do not: a MoA model and a boolean prompt guard.

Models you can call

deepseek-flash: DeepSeek V4.1 Flash direct, $0.11 in / $0.23 out per million tokens.
deepseek-flash-three-pass: Sansi, $0.30 in / $0.60 out per million user-visible tokens. Internal review and revision rounds are not billed.
xjp-guard-1: the prompt guard at /v1/workflows/guard, billed by estimated tokens.

Why a MoA model in a gateway

A gateway can do more than forward requests. Sansi runs three rounds behind one call, with gates that send empty or unparseable output back with the error position, so you get one answer and pay for the tokens you can see.

Start

See /docs/api for a working request, and /pricing to request an API key.

base_url = https://router.xiaojins.com/v1