An OpenRouter alternative with a MoA model and a prompt guard

Where we are more expensive

On 2026-10-07 OpenRouter listed DeepSeek V4 Flash at $0.09 in / $0.19 out per million tokens. Our direct price is $0.11 / $0.23, about 20% above that. If you only want the plain model at the lowest price, that is not us.

What you get for it

• Sansi: the same DeepSeek with a draft, review, revise loop and mechanical gates. On 7 real editing tickets x 8 runs, first-pass success was 50/56 against 41/56 for DeepSeek direct (p=0.026), at about 1.8x median time and spend. Billed at $0.30 / $0.60 per million user-visible tokens.
• A prompt guard that returns only booleans and refuses when it cannot decide.
• A published tuning log: when a new base model ships we rerun the same benchmark and say so on the page.

Switching

The API is OpenAI-compatible, so moving means changing base_url and the key. See /docs/api.