Founded 2026 · Seoul, South Korea

About Seonin

Seonin is a personal AI studio, founded in 2026 and operated by Minjun Gu in Seoul. We build and ship autonomous multi-agent AI products on Anthropic Claude — software where agents plan, generate, and verify, and a human approves every release.

Company facts

Who we are.

CompanySeonin — personal AI studio
Founded2026 (Seoul, South Korea)
Operator / FounderMinjun Gu (구민준)
Domainseonin.online (registered via Cloudflare, valid to 2027)
Company email[email protected] · [email protected] (on-domain, live MX)
HostingCloudflare Pages + Pages Functions (serverless), KV storage
Core modelAnthropic Claude (Opus / Sonnet / Haiku) — production agent pipelines
What we build on Claude

Live products, in production.

These are not planned — they are shipped and running today. Claude drives the reasoning of every pipeline.

In production now

Where each product stands.

Comic StudioEpisode 151 in production — pages 16–24 lettered and shipping, pages 25–26 in the QA loop (1 panel cleared 4/4 on pass 2 in the record above; failing prompts are rewritten by Claude, then re-rendered on a fresh seed).
Novel EngineChapter 3 approved after seven documented passes (10,471 characters, 39% dialogue, restricted-term scan clear) — reverse design, voice lock, continuity ledger, human gate.
Public API

Verifiable, live endpoints.

Seonin exposes a public, documented API. You can verify real usage right now:

GET https://seonin.online/api/view → real-time product usage counters (KV-backed):

{"site":0,"comics":0,"novel":0}

Also: POST /api/view?key=… (increment), POST /api/access (early-access queue with 6-digit verification), GET /api/admin (authenticated). Full endpoint map in /llms.txt.

Technology

How it works.

Anthropic Claude (Opus / Sonnet / Haiku) is the reasoning core across all three products. Seonin runs multi-agent pipelines on Cloudflare Pages + Functions with Workers KV persistence, calls Claude for planning, prompt-writing, and structured judgment, and layers open image/diffusion models (ComfyUI) for rendering. Human-in-the-loop QA gates every published release.