lidge-jun · theQuert · Aug 3, 2026 · Aug 3, 2026 · Aug 4, 2026
@@ -216,7 +216,7 @@ selectors, then retry. Signing in from a machine with no existing `kiro-cli` ses
 
 ## 3. API-key catalog
 
-opencodex ships 69 built-in presets: 58 key-based, seven OAuth, three local, and one default
+opencodex ships 70 built-in presets: 59 key-based, seven OAuth, three local, and one default
 ChatGPT-forward preset. The dashboard's **Add provider** picker opens a key provider's dashboard,
 validates the key, and stores it; validation is provider-specific, and Command Code's public
 catalog reports keys as unverifiable. Notable entries:
@@ -249,6 +249,7 @@ free-experimentation model.
 | MiniMax · MiniMax (CN) | `https://api.minimax.io/v1` · `https://api.minimaxi.com/v1` |
 | DeepSeek | `https://api.deepseek.com` |
 | Cerebras | `https://api.cerebras.ai/v1` |
+| Apertis | `https://api.apertis.ai/v1` |
 | DeepInfra | `https://api.deepinfra.com/v1/openai` |
 | Hyperbolic | `https://api.hyperbolic.xyz/v1` |
 | Baseten Model APIs | `https://inference.baseten.co/v1` |
@@ -270,6 +271,12 @@ free-experimentation model.
 | Cloudflare AI Gateway | `https://gateway.ai.cloudflare.com/v1/{account-id}/{gateway}/anthropic` |
 | …and more | opencode zen, Vercel AI Gateway, Venice, NanoGPT, Synthetic, Qianfan, Alibaba, Parallel, ZenMux, LiteLLM |
 
+**Apertis discovery.** Apertis exposes an OpenAI-compatible API at
+[`https://api.apertis.ai/v1`](https://docs.apertis.ai/api/), uses Bearer API keys, and returns a
+key/plan-scoped live catalog from [`GET /v1/models`](https://docs.apertis.ai/api/utilities/models/).
+Do not freeze a model id from the catalog; availability can vary by key type and plan. Apertis's
+public site identifies the operator as STIMA AI LLC and publishes its [Terms of Service](https://apertis.ai/terms).
+
 Most use the `openai-chat` adapter with a bearer key; a few that expose only an Anthropic-compatible
 endpoint (e.g. **Xiaomi MiMo**) use the `anthropic` adapter (`x-api-key`).
 Volcengine Agent Plan uses its native Responses endpoint through `openai-responses`.

@@ -143,7 +143,7 @@ Kiro のログインには Kiro CLI が必要です。Unix では `curl -fsSL ht
 
 ## 3. API キーカタログ
 
-opencodex には組み込みプリセットが 69 個含まれています。キー方式 58、OAuth 7、ローカル 3、
+opencodex には組み込みプリセットが 70 個含まれています。キー方式 59、OAuth 7、ローカル 3、
 デフォルト ChatGPT 転送プリセット 1 です。ダッシュボードの **Add provider** ピッカーはキー発行ページを開き、
 入力したキーを検証した後保存します(検証はプロバイダー固有で、Command Code の公開カタログはキーを
 検証不能として報告します)。主な項目は以下のとおりです:
@@ -176,6 +176,7 @@ Cline IDE/CLI のみで API からは使えません。`minimax/minimax-m2.5`
 | MiniMax · MiniMax (CN) | `https://api.minimax.io/v1` · `https://api.minimaxi.com/v1` |
 | DeepSeek | `https://api.deepseek.com` |
 | Cerebras | `https://api.cerebras.ai/v1` |
+| Apertis | `https://api.apertis.ai/v1` |
 | DeepInfra | `https://api.deepinfra.com/v1/openai` |
 | Hyperbolic | `https://api.hyperbolic.xyz/v1` |
 | Baseten Model APIs | `https://inference.baseten.co/v1` |

@@ -142,7 +142,7 @@ Kiro 로그인에는 Kiro CLI가 필요합니다. Unix에서는 `curl -fsSL http
 
 ## 3. API 키 카탈로그
 
-opencodex에는 빌트인 프리셋이 69개 들어 있습니다. 키 방식 58개, OAuth 7개, 로컬 3개,
+opencodex에는 빌트인 프리셋이 70개 들어 있습니다. 키 방식 59개, OAuth 7개, 로컬 3개,
 기본 ChatGPT 포워드 프리셋 1개입니다. 대시보드의 **Add provider** 선택기는 키 발급 페이지를 열고,
 입력한 키를 검증한 뒤 저장합니다(검증은 프로바이더별로 다르며, Command Code의 공개 카탈로그는 키를
 검증 불가로 보고합니다). 주요 항목은 다음과 같습니다:
@@ -176,6 +176,7 @@ Cline IDE/CLI에서만 제공되며 API로는 사용할 수 없습니다. `minim
 | MiniMax · MiniMax (CN) | `https://api.minimax.io/v1` · `https://api.minimaxi.com/v1` |
 | DeepSeek | `https://api.deepseek.com` |
 | Cerebras | `https://api.cerebras.ai/v1` |
+| Apertis | `https://api.apertis.ai/v1` |
 | DeepInfra | `https://api.deepinfra.com/v1/openai` |
 | Hyperbolic | `https://api.hyperbolic.xyz/v1` |
 | Baseten Model APIs | `https://inference.baseten.co/v1` |

@@ -153,7 +153,7 @@ OAuth-провайдеры, чьи учётные данные содержат
 
 ## 3. Каталог API-ключей
 
-opencodex поставляется с 69 встроенными пресетами: 58 на основе ключей, семь OAuth, три локальных и
+opencodex поставляется с 70 встроенными пресетами: 59 на основе ключей, семь OAuth, три локальных и
 один пресет ChatGPT-форварда по умолчанию. Селектор **Add provider** в дашборде открывает страницу
 выдачи ключей провайдера, проверяет ключ и сохраняет его; проверка зависит от провайдера, а публичный
 каталог Command Code сообщает ключ как непроверенный. Наиболее заметные записи:
@@ -186,6 +186,7 @@ opencodex поставляется с 69 встроенными пресетам
 | MiniMax · MiniMax (CN) | `https://api.minimax.io/v1` · `https://api.minimaxi.com/v1` |
 | DeepSeek | `https://api.deepseek.com` |
 | Cerebras | `https://api.cerebras.ai/v1` |
+| Apertis | `https://api.apertis.ai/v1` |
 | DeepInfra | `https://api.deepinfra.com/v1/openai` |
 | Hyperbolic | `https://api.hyperbolic.xyz/v1` |
 | Baseten Model APIs | `https://inference.baseten.co/v1` |

@@ -131,7 +131,7 @@ Kiro 登录需要 Kiro CLI：Unix 使用 `curl -fsSL https://cli.kiro.dev/instal
 
 ## 3. API 密钥目录
 
-opencodex 内置 69 个预设：58 个密钥预设、7 个 OAuth 预设、3 个本地预设，以及 1 个默认的
+opencodex 内置 70 个预设：59 个密钥预设、7 个 OAuth 预设、3 个本地预设，以及 1 个默认的
 ChatGPT 转发预设。仪表盘的 **Add provider** 选择器会打开密钥提供商的控制台，验证并保存密钥。
 验证因提供商而异，Command Code 的公开目录会将密钥报告为无法验证。主要条目包括：
 
@@ -163,6 +163,7 @@ Cline IDE/CLI 中提供，不能通过 API 使用；`minimax/minimax-m2.5` 是
 | MiniMax · MiniMax (CN) | `https://api.minimax.io/v1` · `https://api.minimaxi.com/v1` |
 | DeepSeek | `https://api.deepseek.com` |
 | Cerebras | `https://api.cerebras.ai/v1` |
+| Apertis | `https://api.apertis.ai/v1` |
 | DeepInfra | `https://api.deepinfra.com/v1/openai` |
 | Hyperbolic | `https://api.hyperbolic.xyz/v1` |
 | Baseten Model APIs | `https://inference.baseten.co/v1` |

@@ -6,11 +6,11 @@
   "scripts": {
     "dev": "vite",
     "build": "tsc -b && vite build",
-    "lint": "eslint .",
+    "lint": "bun --bun eslint .",
     "test": "bun test tests",
-    "lint:i18n": "eslint src/pages src/components src/App.tsx src/ui.tsx",
-    "doctor": "npx --yes react-doctor@0.9.3 --verbose --scope changed --base origin/main --no-telemetry",
-    "doctor:full": "npx --yes react-doctor@0.9.3 --verbose --scope full --no-telemetry",
+    "lint:i18n": "bun --bun eslint src/pages src/components src/App.tsx src/ui.tsx",
+    "doctor": "npm exec --yes --package=typescript@6.0.2 --package=react-doctor@0.9.3 -- react-doctor --verbose --scope changed --base origin/main --no-telemetry",
+    "doctor:full": "npm exec --yes --package=typescript@6.0.2 --package=react-doctor@0.9.3 -- react-doctor --verbose --scope full --no-telemetry",
     "preview": "vite preview"
   },
   "dependencies": {

@@ -71,6 +71,7 @@ const PROVIDER_DISPLAY_NAMES: Record<string, string> = {
   xiaomi: "Xiaomi",
   cursor: "Cursor",
   deepseek: "DeepSeek",
+  apertis: "Apertis",
   github: "GitHub",
   "github-copilot": "GitHub Copilot",
   "gitlab-duo": "GitLab Duo",

@@ -0,0 +1,2 @@
+schema: spec-driven
+created: 2026-08-03
@@ -0,0 +1,4 @@
+# apertis-reference-directory
+
+Restore Apertis as a canonical OpenAI-compatible API-key provider with authenticated live model
+discovery, while keeping the maintainer evidence and security-review boundary explicit.
@@ -0,0 +1,84 @@
+## Context
+
+Canonical registry entries seed provider configuration, expose API-key setup, enrich saved
+providers, and feed routing and the shared Codex catalog. Apertis's official API documentation
+currently describes the OpenAI-compatible base URL `https://api.apertis.ai/v1`, Bearer
+authentication, and an authenticated `/v1/models` endpoint whose result depends on the API key
+type/plan. Apertis's public website identifies STIMA AI LLC as the operator and describes
+multi-provider routing.
+
+The upstream repository requires additional private routing/resale authorization and maintainer
+security review before a credential-destination preset can merge. This change therefore implements
+the requested canonical behavior and records that external gate honestly; it does not infer private
+authorization from public product or API documentation.
+
+## Goals / Non-Goals
+
+**Goals:**
+
+- Register Apertis as a canonical `authKind: "key"`, `adapter: "openai-chat"` provider.
+- Seed the documented fixed base URL and dashboard URL into CLI/key-login/dashboard flows.
+- Validate keys and discover model ids through the authenticated OpenAI-shaped `/models` path.
+- Route chat-completion requests to the canonical host while preserving same-named custom
+  providers' own adapter, destination, and credentials.
+- Keep model availability live and key-scoped rather than encoding a speculative static catalog.
+- Keep public evidence links and the private maintainer authorization gate separate.
+
+**Non-Goals:**
+
+- Add a new adapter, Responses wire, image/audio integration, or provider-specific request
+  semantics beyond the existing OpenAI Chat adapter.
+- Freeze model ids, context windows, prices, or a default model from the mutable live catalog.
+- Claim that public endpoint/terms/legal-entity evidence proves private routing/resale authorization.
+- Send credentials, post maintainer messages, push the branch, merge, or deploy as part of local
+  implementation.
+
+## Decisions
+
+- **Extend the existing registry-derived provider flow.** This keeps CLI init, key-login,
+  dashboard presets, routing, and catalog derivation on one source of truth instead of adding an
+  Apertis-specific path.
+- **Use live discovery with the default `/models` resolution.** The configured base URL already
+  ends in `/v1`, so the existing discovery and key-validation flow produces `/v1/models`, sends a
+  Bearer key, and applies the repository's bounded response/row guards. A registry-specific
+  discovery policy is unnecessary for the documented endpoint.
+- **Set `preserveCustomDestination: true`.** A saved provider named `apertis` with a custom
+  adapter/base URL must not be silently retargeted when the canonical preset is introduced.
+- **Do not add static models or a default model.** Official documentation says availability varies
+  by key type and plan; live discovery is the faithful contract and avoids stale claims.
+- **Use the existing GUI display-name classification without inventing an icon asset.** The provider
+  remains discoverable and correctly labeled while the default unknown-provider icon fallback stays
+  intact.
+- **Keep the evidence gate outside runtime behavior.** Public API/terms/legal-entity links are
+  recorded in source/docs; private routing/resale authorization remains a maintainer review input.
+- **Keep GUI validation runtimes explicit.** ESLint continues to use Bun's runtime bridge, while
+  React Doctor uses npm exec with pinned TypeScript and React Doctor packages so Bun cannot omit
+  React Doctor's TypeScript peer.
+
+## Risks / Trade-offs
+
+- [A user key and traffic reach a multi-provider gateway] → fixed-host routing, explicit public
+  provenance, `preserveCustomDestination`, and maintainer/security review are required before merge.
+- [The live catalog changes by key/plan] → no static model list/default is claimed; discovery is
+  bounded and fixture-tested.
+- [A same-named custom provider could be disabled in GUI classification] → the canonical entry's
+  destination-preservation contract and regression test keep custom adapter/base URL controls intact.
+- [Public docs may be mistaken for private authorization] → the change records public evidence and
+  the unresolved routing/resale gate separately.
+- [A GUI validation command could lose its local TypeScript peer under Bun] → ESLint scripts use
+  Bun's runtime bridge and React Doctor scripts use npm exec with explicit TypeScript and React
+  Doctor packages; direct external invocations remain outside this repository contract.
+
+## Migration Plan
+
+Fresh Apertis setup derives the canonical fixed destination. Existing saved `apertis` providers
+retain their configured adapter, base URL, and key when they do not match the canonical transport.
+Rolling back removes the preset surfaces but does not rewrite or delete existing provider config;
+explicit custom routes remain user-owned.
+
+## Open Questions
+
+- Which maintainer-designated private channel will record the routing/resale authorization, named
+  maintenance owner, and verification date required for merge?
+- After the current head is reviewed, does the maintainer require any provider-specific capability
+  exclusions beyond the existing `openai-chat` contract?
@@ -0,0 +1,46 @@
+## Why
+
+Apertis publishes an OpenAI-compatible API at `https://api.apertis.ai/v1`, uses Bearer API keys,
+and exposes a key-scoped live `/v1/models` catalog. The current PR intentionally removed the
+canonical entry and therefore no longer meets the requested outcome of adding Apertis as a model
+provider.
+
+## What Changes
+
+- Restore Apertis as a canonical API-key provider in the registry-derived CLI, dashboard, key-login,
+  routing, and model-discovery surfaces.
+- Use the documented fixed base URL and authenticated live model discovery without freezing a
+  mutable model list or claiming a default model.
+- Preserve same-named custom providers' adapter, destination, and key boundary.
+- Add focused coverage for derivation, Bearer validation, live discovery, chat routing, and custom
+  provider preservation.
+- Sync the provider catalog documentation and translated tables/counts.
+- Retain the existing Bun-based GUI lint runtime fix and make React Doctor's TypeScript peer
+  resolution explicit for Bun-launched validation.
+- Keep the maintainer-required routing/resale authorization and security review as an explicit
+  pre-merge gate; public endpoint/terms/legal-entity evidence must not be presented as proof of
+  private authorization.
+
+## Capabilities
+
+### New Capabilities
+
+- `provider-canonical-integration`: expose Apertis as a canonical key provider with authenticated
+  live model discovery and fixed-host routing.
+- `gui-lint-runtime`: execute GUI lint and React Doctor validation with runtimes that preserve
+  the repository's local TypeScript tooling.
+
+### Modified Capabilities
+
+<!-- None. The previous reference-only contract is replaced by the canonical integration capability
+in this still-unmerged change. -->
+
+## Impact
+
+- `src/providers/registry.ts`, registry derivation, key validation, GUI provider classification,
+  and provider parity gain the `apertis` canonical entry.
+- `tests/apertis-provider.test.ts`, the discovery fixture, and registry parity tests prove the
+  provider contract without contacting the live service.
+- English and translated `docs-site` provider guides list the new endpoint and updated counts.
+- `gui/package.json` lint commands continue to use Bun's runtime bridge, while React Doctor
+  commands use npm's explicit package resolution instead of a Bun-translated `npx` launch.
@@ -0,0 +1,28 @@
+## ADDED Requirements
+
+### Requirement: GUI lint scripts load the local TypeScript ESLint plugin
+The GUI package's `lint` and `lint:i18n` scripts SHALL execute ESLint through Bun's runtime bridge
+so that the checked-in ESLint configuration and its local TypeScript plugin load successfully.
+
+#### Scenario: Full GUI lint command
+- **WHEN** a contributor runs the GUI `lint` script from the repository validation path
+- **THEN** ESLint evaluates the configured GUI files without an unknown-TypeScript-extension error
+
+#### Scenario: Focused GUI i18n lint command
+- **WHEN** a contributor runs the GUI `lint:i18n` script
+- **THEN** ESLint evaluates the configured UI paths with the same local plugin available
+
+### Requirement: React Doctor scripts resolve their TypeScript peer explicitly
+The GUI package's `doctor` and `doctor:full` scripts SHALL invoke React Doctor through npm
+with the pinned TypeScript and React Doctor packages so that Bun's `npx` alias cannot omit the
+TypeScript peer required by React Doctor.
+
+#### Scenario: Changed GUI React Doctor command
+- **WHEN** a contributor runs the GUI `doctor` script from the repository validation path
+- **THEN** React Doctor evaluates the changed GUI scope and resolves its TypeScript peer
+  successfully
+
+#### Scenario: Full GUI React Doctor command
+- **WHEN** a contributor runs the GUI `doctor:full` script
+- **THEN** React Doctor evaluates the full GUI scope with the same explicit TypeScript peer
+  resolution