Update README.md

README.md

@@ -11,83 +11,43 @@ pinned: false
 
 **A model lab for white-hat security. / ホワイトハッカーのためのモデルラボ。**
 
-Kotoba Labs publishes open-weight language models tuned for **authorized security
-research**. Our models power [**Kotoba Cloud**](https://kotoba.cloud) — *"Empower those
-who protect"* — and are part of the [**Kotoba**](https://kotoba-lang.org) ecosystem for
-safe, AI-generated software.
+Kotoba Labs publishes open-weight language models tuned for **authorized security research**. Our models power [**Kotoba Cloud**](https://kotoba.cloud) — *"Empower those who protect"* — and are part of the [**Kotoba**](https://kotoba-lang.org) ecosystem for safe, AI-generated software.
 
-Kotoba Labs は、**認可されたセキュリティ研究**のためのオープンウェイト言語モデルを公開して
-います。これらのモデルは [**Kotoba Cloud**](https://kotoba.cloud) を支え、安全な
-AI 生成ソフトウェアのエコシステム [**Kotoba**](https://kotoba-lang.org) の一部です。
+Kotoba Labs は、**認可されたセキュリティ研究**のためのオープンウェイト言語モデルを公開しています。これらのモデルは [**Kotoba Cloud**](https://kotoba.cloud) を支え、安全な AI 生成ソフトウェアのエコシステム [**Kotoba**](https://kotoba-lang.org) の一部です。
 
 ---
 
 ## 目的 / Purpose
 
-- **Defensive by intent.** We build models that help code auditors and white-hat
-  researchers review code, validate vulnerabilities, and prepare fixes — the work
-  [Kotoba Cloud](https://kotoba.cloud) exists to support.
-- **Honest, reproducible builds.** Each model documents exactly what was changed, the
-  method, the base model, and the serving footprint. No hidden capabilities, no
-  marketing claims we can't reproduce.
-- **Hardware-real optimization.** Weights are quantized to the format the target GPU
-  actually runs — e.g. **NVFP4** for NVIDIA **RTX PRO 6000** (Blackwell) — so a published
-  model maps to a concrete, runnable deployment.
-
-コード監査者とホワイトハッカーが、コードのレビュー・脆弱性の検証・修正の準備を行う作業を
-支援するモデルを、**再現可能な形**で公開します。改変内容・手法・ベースモデル・必要 VRAM を
-すべて明記します。
-
----
-
-## 収録モデル / What we publish
-
-- **White-hat (abliterated) builds** — refusal-direction-ablated derivatives of
-  permissively licensed open-weight models, so an authorized researcher is not blocked
-  mid-investigation.
-- **RTX PRO 6000-optimized weights** — NVFP4 / FP8 builds sized for single- or
-  multi-card Blackwell serving via **SGLang** / **vLLM**.
+- **Defensive by intent.** We build models that help code auditors and white-hat researchers review code, validate vulnerabilities, and prepare fixes — the work [Kotoba Cloud](https://kotoba.cloud) exists to support.
+- **Honest, reproducible builds.** Each model documents exactly what was changed, the method, the base model, and the serving footprint. No hidden capabilities, no marketing claims we can't reproduce.
+- **Hardware-real optimization.** Weights are quantized to the format the target GPU actually runs — e.g. **NVFP4** for NVIDIA **RTX PRO 6000** (Blackwell) — so a published model maps to a concrete, runnable deployment.
 
 ---
 
-## 規約 / Terms
+## Kotoba Cloud — what these models power
 
-- **Licensing.** Each model is a derivative and **inherits the license of its base model**
-  (e.g. Apache-2.0). The base model's terms apply to your use; check each model card.
-- **Attribution.** Base models are credited on every model card with a link.
-- **As-is.** Models are provided without warranty. You are responsible for validating
-  outputs before relying on them.
+[Kotoba Cloud](https://kotoba.cloud) is where these weights are put to work: a platform for identity-verified, authorized security researchers, walking a fixed workflow — **INVESTIGATE → VALIDATE → REMEDIATE** — from inspecting in-scope code, to confirming a finding, to preparing a responsible-disclosure fix. *"One person's discovery becomes a community's strength."*
 
-各モデルはベースモデルの派生物であり、**ベースモデルのライセンス（例: Apache-2.0）を継承**
-します。利用時はベースモデルの規約が適用されます。
+- **Access control**: Passkey + post-quantum (ML-DSA-65) authentication, Stripe Identity eKYC, AML/CTF screening, and per-engagement scope approval gate every model call.
+- **Free tier for verified researchers**: 1,000 requests/day, up to 32,768 tokens.
+- **Verifiable supply chain**: releases are signed (Ed25519 + ML-DSA-65) and distributed over content-addressed IPFS/CID, backed by an IPLD graph database.
 
----
+Kotoba Cloud は、本人確認済みの認可セキュリティ研究者向けプラットフォームです。**調査 (INVESTIGATE) → 検証 (VALIDATE) → 修復 (REMEDIATE)** という固定ワークフローで、対象コード��検査から発見の妥当性確認、責任ある開示に向けた修正準備までを支えます。「一人の発見がコミュニティの力になる」がその理念です。
 
-## 制限・許容利用方針 / Restrictions & Acceptable Use
+## Models / モデル
 
-Several of our models are **abliterated** ("uncensored"): a safety layer has been removed
-so the model does not reflexively refuse. This is **dual-use**.
+Browse our published weights above in the Models tab — security-research-tuned fine-tunes and quantized builds matched to real inference hardware.
 
-一部のモデルは **abliterated（uncensored）** です。安全層が除去されており、これは
-**デュアルユース**です。以下を順守してください。
-
-- ✅ **Permitted:** authorized security research and red-teaming, testing **your own**
-  systems, sanctioned engagements, CTFs, model-safety evaluation, education.
-- ⛔ **Not permitted:** any unlawful activity; targeting systems, data, or people you are
-  not authorized to test; producing weapons, malware, or fraud for real-world harm;
-  harassment or targeting of private individuals.
-- **You are responsible** for how you use these models and for staying within an
-  authorized scope and applicable law. Removing a refusal filter does not remove your
-  legal and ethical obligations.
+## Links
 
-認可された範囲・適用法令の遵守は利用者の責任です。拒否フィルタの除去は、法的・倫理的責任を
-免除するものではありません。
+- 🌐 Platform: [kotoba.cloud](https://kotoba.cloud)
+- 📖 Docs: [docs.kotoba.cloud](https://docs.kotoba.cloud)
+- 🧩 Language: [kotoba-lang.org](https://kotoba-lang.org)
+- 🗂️ Graph DB: [graph.kotoba.cloud](https://graph.kotoba.cloud)
+- 🤝 Agent work: [itonami.cloud](https://itonami.cloud)
+- ✉️ Support: support@kotoba.cloud
 
 ---
 
-## Links
-
-- 🛡️ **Kotoba Cloud** — white-hat security inference · <https://kotoba.cloud>
-- ⚙️ **Kotoba Language** — a safe, fast language for AI-generated software · <https://kotoba-lang.org>
-
-*Kotoba: "grant narrowly, then compile."*
+*Use of Kotoba Cloud and these models is restricted to authorized security research on in-scope targets. See [kotoba.cloud](https://kotoba.cloud) for details.*