egaki 0.2.0 → 0.4.0

package/src/login.ts

@@ -1,6 +1,7 @@
 // Login command — interactive and non-interactive API key management.
 // Interactive mode uses clack prompts for provider selection and key input.
 // Non-interactive mode supports --provider + --key flags and stdin piping.
+// ChatGPT OAuth uses a browser flow instead of pasting a key.
 import {
   intro,
   outro,
@@ -17,8 +18,10 @@ import {
   saveProviderKey,
   removeProviderKey,
   getKeyStatus,
-  readCredentials,
+  getChatGptAuth,
+  saveChatGptAuth,
 } from './credentials.js'
+import { chatGptOAuthLogin, extractPlanType } from './chatgpt-auth.js'
 
 export async function loginInteractive(): Promise<void> {
   intro(pc.bold('egaki login'))
@@ -32,6 +35,11 @@ export async function loginInteractive(): Promise<void> {
       if (status.source === 'stored') {
         return pc.green('(saved)')
       }
+      if (status.source === 'oauth') {
+        const auth = getChatGptAuth()
+        const email = auth?.email
+        return pc.green(`(signed in${email ? ` as ${email}` : ''})`)
+      }
       return pc.dim('(not configured)')
     })()
     // Highlight the egaki option as recommended
@@ -61,6 +69,14 @@ export async function loginInteractive(): Promise<void> {
     process.exit(1)
   }
 
+  // ChatGPT uses browser OAuth instead of key paste
+  if (provider === 'chatgpt') {
+    const auth = await chatGptOAuthLogin()
+    saveChatGptAuth(auth)
+    outro('Done — ChatGPT OAuth saved')
+    return
+  }
+
   log.info(
     `${pc.dim('Env var:')} ${info.envVar} ${pc.dim('(also used by the CLI when set)')}`,
   )
@@ -91,13 +107,13 @@ export async function loginInteractive(): Promise<void> {
   outro('Done')
 }
 
-export function loginNonInteractive({
+export async function loginNonInteractive({
   provider,
   key,
 }: {
   provider: string
   key: string
-}): void {
+}): Promise<void> {
   const info = PROVIDERS[provider]
   if (!info) {
     const available = Object.keys(PROVIDERS).join(', ')
@@ -107,6 +123,14 @@ export function loginNonInteractive({
     process.exit(1)
   }
 
+  // ChatGPT uses browser OAuth — key flag is not needed
+  if (provider === 'chatgpt') {
+    const auth = await chatGptOAuthLogin()
+    saveChatGptAuth(auth)
+    console.log(pc.green('ChatGPT OAuth saved'))
+    return
+  }
+
   if (!key || key.trim().length === 0) {
     console.error(pc.red('API key cannot be empty'))
     process.exit(1)
@@ -121,6 +145,26 @@ export function showLoginStatus(): void {
 
   for (const [key, info] of Object.entries(PROVIDERS)) {
     const status = getKeyStatus(key)
+
+    if (key === 'chatgpt') {
+      const auth = getChatGptAuth()
+      if (auth) {
+        const plan = extractPlanType(auth)
+        const email = auth.email ?? auth.accountId ?? 'unknown'
+        const expired = auth.expires < Date.now()
+        const expiryLabel = expired
+          ? pc.yellow('(token expired, will auto-refresh)')
+          : pc.dim(`(token valid)`)
+        console.log(`${pc.green('*')} ${info.label} ${pc.green('(signed in)')}`)
+        console.log(pc.dim(`  account: ${email}${plan ? `, plan: ${plan}` : ''}`))
+        console.log(`  ${expiryLabel}`)
+      } else {
+        console.log(`${pc.dim('-')} ${info.label} ${pc.dim('(not signed in)')}`)
+        console.log(pc.dim(`  run: egaki login → select ChatGPT`))
+      }
+      continue
+    }
+
     const icon = status.available ? pc.green('*') : pc.dim('-')
     const source = (() => {
       if (status.source === 'env') {
@@ -150,7 +194,8 @@ export function removeLogin(provider: string): void {
   }
 
   removeProviderKey(provider)
-  console.log(pc.green(`${info.label} key removed`))
+  const label = provider === 'chatgpt' ? 'ChatGPT OAuth' : info.label
+  console.log(pc.green(`${label} key removed`))
 }
 
 // Read API key from stdin (for piping: echo "sk-xxx" | egaki login --provider google)

package/src/model-catalog.ts

@@ -109,6 +109,22 @@ const xaiImage = {
   strategy: 'image' as const,
 }
 
+const vertexImagen = {
+  provider: 'vertex',
+  strategy: 'image' as const,
+  features: {
+    ...googleImagen.features,
+  },
+}
+
+const vertexText = {
+  provider: 'vertex',
+  strategy: 'text' as const,
+  features: {
+    ...googleText.features,
+  },
+}
+
 const fluxAspectRatios = ['1:1', '3:4', '4:3', '9:16', '16:9', '9:21', '21:9']
 
 // ─── catalog ─────────────────────────────────────────────────────────────────
@@ -196,6 +212,64 @@ export const CATALOG: ModelEntry[] = [
     cost: { type: 'per-token', inputPerM: 0.5, outputPerM: 3.0 },
   },
 
+  // ── Vertex: Imagen ──────────────────────────────────────────────────────
+  // Same models as Google AI Studio but routed through Vertex AI / Google Cloud billing.
+  // Use `egaki login --provider vertex --key <key>` to configure.
+  {
+    id: 'vertex/imagen-4.0-generate-001',
+    name: 'Imagen 4 (Vertex)',
+    released: '2025-08-15',
+    ...vertexImagen,
+    cost: { type: 'per-image', perImage: 0.04 },
+  },
+  {
+    id: 'vertex/imagen-4.0-ultra-generate-001',
+    name: 'Imagen 4 Ultra (Vertex)',
+    released: '2025-08-15',
+    ...vertexImagen,
+    cost: { type: 'per-image', perImage: 0.06 },
+  },
+  {
+    id: 'vertex/imagen-4.0-fast-generate-001',
+    name: 'Imagen 4 Fast (Vertex)',
+    released: '2025-08-15',
+    ...vertexImagen,
+    cost: { type: 'per-image', perImage: 0.02 },
+  },
+
+  // ── Vertex: Gemini text+image ─────────────────────────────────────────
+  {
+    id: 'vertex/gemini-2.5-flash-image',
+    name: 'Gemini 2.5 Flash Image (Vertex)',
+    released: '2025-08-26',
+    ...vertexText,
+    cost: { type: 'per-token', inputPerM: 0.3, outputPerM: 30 },
+  },
+  {
+    id: 'vertex/gemini-3-pro-image-preview',
+    name: 'Gemini 3 Pro Image (Vertex)',
+    released: '2025-11-20',
+    ...vertexText,
+    cost: { type: 'per-token', inputPerM: 1.25, outputPerM: 10 },
+  },
+  {
+    id: 'vertex/gemini-3.1-flash-image-preview',
+    name: 'Gemini 3.1 Flash Image (Vertex)',
+    released: '2026-02-26',
+    ...vertexText,
+    features: {
+      ...vertexText.features,
+      aspectRatios: [
+        ...vertexText.features.aspectRatios,
+        '4:1',
+        '1:4',
+        '8:1',
+        '1:8',
+      ],
+    },
+    cost: { type: 'per-token', inputPerM: 0.5, outputPerM: 3.0 },
+  },
+
   // ── BFL (AI Gateway) ───────────────────────────────────────────────────
   {
     id: 'flux-kontext-max',
@@ -432,9 +506,24 @@ export const CATALOG: ModelEntry[] = [
       multipleImages: true,
     },
   },
-  // NOTE: chatgpt-image-latest exists in OpenAI's .d.ts types but it's a moving
-  // alias used by ChatGPT/Responses API, not a stable image endpoint model.
-  // Omitted until OpenAI stabilizes it as a first-class image generation target.
+  {
+    id: 'chatgpt-image-latest',
+    name: 'ChatGPT Image',
+    description:
+      'Rolling-latest alias that tracks the newest ChatGPT image model. Currently points ' +
+      'to gpt-image-1.5. Pricing may shift when the alias retargets to a newer model.',
+    released: '2026-01',
+    ...openaiImage,
+    cost: { type: 'per-image', perImage: 0.034 },
+    features: {
+      editing: true,
+      inpainting: true,
+      aspectRatios: [],
+      sizes: ['1024x1024', '1536x1024', '1024x1536'],
+      seed: false,
+      multipleImages: true,
+    },
+  },
 
   // ── Replicate ──────────────────────────────────────────────────────────
   {

package/src/models.ts

@@ -17,7 +17,7 @@
 // dist/index.d.ts and add entries here.
 import type { ImageModel, LanguageModel } from 'ai'
 import pc from 'picocolors'
-import { PROVIDERS, EGAKI_GATEWAY_URL } from './credentials.js'
+import { PROVIDERS, EGAKI_GATEWAY_URL, shouldUseChatGptBackend } from './credentials.js'
 import { CATALOG, findModel } from './model-catalog.js'
 import type { ModelEntry } from './model-catalog.js'
 import { VIDEO_CATALOG, findVideoModel } from './video-model-catalog.js'
@@ -29,6 +29,16 @@ export type { ModelEntry, VideoModelEntry }
 export const IMAGE_MODELS = CATALOG.map((m) => m.id) as [string, ...string[]]
 export const VIDEO_MODELS = VIDEO_CATALOG.map((m) => m.id) as [string, ...string[]]
 
+/**
+ * Strip provider prefix from a model ID (e.g. "vertex/imagen-4.0-generate-001" → "imagen-4.0-generate-001").
+ * Provider SDKs expect bare model IDs, but the catalog uses prefixed IDs for routing.
+ */
+function stripProviderPrefix(modelId: string): string {
+  const slash = modelId.indexOf('/')
+  if (slash === -1) return modelId
+  return modelId.slice(slash + 1)
+}
+
 export const DEFAULT_MODEL = 'nano-banana-pro-preview'
 export const DEFAULT_VIDEO_MODEL = 'veo-3.1-fast-generate-001'
 
@@ -66,6 +76,20 @@ export function ensureProviderKey(providerName: string): void {
   // If direct provider key exists, we're good
   if (hasDirectProviderKey(providerName)) return
 
+  // Vertex models require a direct key — the upstream Vercel AI Gateway
+  // does not support vertex/ routing, so egaki gateway can't proxy them.
+  if (providerName === 'vertex') {
+    console.error('')
+    console.error(pc.red(pc.bold('Vertex models require a direct Google Cloud API key')))
+    console.error('')
+    console.error(`  ${pc.cyan('egaki login --provider vertex --key <key>')}`)
+    console.error('')
+    console.error(pc.dim('Get a key at https://console.cloud.google.com/apis/credentials'))
+    console.error(pc.dim('Egaki subscription does not cover Vertex routing yet.'))
+    console.error('')
+    process.exit(1)
+  }
+
   // If egaki key exists, we'll route through the gateway
   if (hasEgakiKey()) return
 
@@ -109,8 +133,11 @@ async function createGatewayImageModel(modelId: string, provider: string): Promi
     apiKey: process.env['EGAKI_API_KEY']!,
     baseURL: EGAKI_GATEWAY_URL,
   })
-  // Gateway expects provider/model format for routing
-  const gatewayModelId = `${provider}/${modelId}`
+  // Gateway expects provider/model format for routing.
+  // For models that already have a prefix (e.g. vertex/imagen-4.0), strip it
+  // and use the catalog provider. For bare IDs, prepend the provider.
+  const bareId = stripProviderPrefix(modelId)
+  const gatewayModelId = `${provider}/${bareId}`
   return gateway.image(gatewayModelId)
 }
 
@@ -123,7 +150,8 @@ async function createGatewayTextModel(modelId: string, provider: string): Promis
     apiKey: process.env['EGAKI_API_KEY']!,
     baseURL: EGAKI_GATEWAY_URL,
   })
-  const gatewayModelId = `${provider}/${modelId}`
+  const bareId = stripProviderPrefix(modelId)
+  const gatewayModelId = `${provider}/${bareId}`
   return gateway(gatewayModelId)
 }
 
@@ -136,10 +164,21 @@ async function createGatewayVideoModel(modelId: string, provider: string) {
     apiKey: process.env['EGAKI_API_KEY']!,
     baseURL: EGAKI_GATEWAY_URL,
   })
-  const gatewayModelId = `${provider}/${modelId}`
+  const bareId = stripProviderPrefix(modelId)
+  const gatewayModelId = `${provider}/${bareId}`
   return gateway.video(gatewayModelId)
 }
 
+/**
+ * ChatGPT OAuth image generation goes through the Codex backend instead of the
+ * OpenAI Image API, so OpenAI image models need the custom responses path.
+ */
+export function shouldUseResponsesApi(modelId: string): boolean {
+  const config = findModel(modelId)
+  if (!config) return false
+  return config.provider === 'openai' && config.strategy === 'image' && shouldUseChatGptBackend()
+}
+
 // Lazily import the provider and create the right model instance.
 // This avoids loading all provider SDKs upfront — only the one needed
 // for the selected model gets imported.
@@ -156,6 +195,10 @@ export async function createImageModel(modelId: string): Promise<ImageModel> {
         const { google } = await import('@ai-sdk/google')
         return google.image(modelId)
       }
+      case 'vertex': {
+        const { vertex } = await import('@ai-sdk/google-vertex')
+        return vertex.image(stripProviderPrefix(modelId))
+      }
       case 'openai': {
         const { openai } = await import('@ai-sdk/openai')
         return openai.image(modelId)
@@ -198,11 +241,15 @@ export async function createTextModel(
         const { google } = await import('@ai-sdk/google')
         return google(modelId)
       }
+      case 'vertex': {
+        const { vertex } = await import('@ai-sdk/google-vertex')
+        return vertex(stripProviderPrefix(modelId))
+      }
       default:
-        // Only Google supports generateText with responseModalities for images
+        // Only Google/Vertex support generateText with responseModalities for images
         console.error(
           pc.red(
-            `Text+image generation is only supported for Google models, got provider: ${config.provider}`,
+            `Text+image generation is only supported for Google/Vertex models, got provider: ${config.provider}`,
           ),
         )
         process.exit(1)
@@ -234,6 +281,10 @@ export async function createVideoModel(modelId: string): Promise<any> {
         const { google } = await import('@ai-sdk/google')
         return google.video(modelId)
       }
+      case 'vertex': {
+        const { vertex } = await import('@ai-sdk/google-vertex')
+        return vertex.video(stripProviderPrefix(modelId))
+      }
       case 'fal': {
         const { fal } = await import('@ai-sdk/fal')
         return fal.video(modelId)
@@ -241,7 +292,7 @@ export async function createVideoModel(modelId: string): Promise<any> {
       default:
         console.error(
           pc.red(
-            `Direct video generation is only supported for Google and Fal keys, got provider: ${config.provider}`,
+            `Direct video generation is only supported for Google, Vertex, and Fal keys, got provider: ${config.provider}`,
           ),
         )
         process.exit(1)

package/src/video-model-catalog.ts

@@ -169,6 +169,67 @@ export const VIDEO_CATALOG: VideoModelEntry[] = [
       multipleVideos: true,
     },
   },
+  // ── Vertex: Veo ──────────────────────────────────────────────────────
+  // Same Veo models routed through Vertex AI / Google Cloud billing.
+  // Use `egaki login --provider vertex --key <key>` to configure.
+  {
+    id: 'vertex/veo-3.1-generate-001',
+    name: 'Veo 3.1 (Vertex)',
+    provider: 'vertex',
+    strategy: 'video',
+    released: '2026-01',
+    cost: {
+      type: 'per-video-second',
+      defaultDurationSec: 8,
+      tiers: [
+        { resolution: '720p', audio: false, costPerSecond: 0.2 },
+        { resolution: '720p', audio: true, costPerSecond: 0.4 },
+        { resolution: '1080p', audio: false, costPerSecond: 0.2 },
+        { resolution: '1080p', audio: true, costPerSecond: 0.4 },
+        { resolution: '4k', audio: false, costPerSecond: 0.4 },
+        { resolution: '4k', audio: true, costPerSecond: 0.6 },
+      ],
+    },
+    features: {
+      textToVideo: true,
+      imageToVideo: false,
+      capabilities: ['t2v'],
+      aspectRatios: ['16:9', '9:16'],
+      resolutions: ['720p', '1080p', '4k'],
+      durationRangeSec: { min: 4, max: 8 },
+      seed: true,
+      multipleVideos: true,
+    },
+  },
+  {
+    id: 'vertex/veo-3.1-fast-generate-001',
+    name: 'Veo 3.1 Fast (Vertex)',
+    provider: 'vertex',
+    strategy: 'video',
+    released: '2026-01',
+    cost: {
+      type: 'per-video-second',
+      defaultDurationSec: 8,
+      tiers: [
+        { resolution: '720p', audio: false, costPerSecond: 0.1 },
+        { resolution: '720p', audio: true, costPerSecond: 0.15 },
+        { resolution: '1080p', audio: false, costPerSecond: 0.1 },
+        { resolution: '1080p', audio: true, costPerSecond: 0.15 },
+        { resolution: '4k', audio: false, costPerSecond: 0.3 },
+        { resolution: '4k', audio: true, costPerSecond: 0.35 },
+      ],
+    },
+    features: {
+      textToVideo: true,
+      imageToVideo: false,
+      capabilities: ['t2v'],
+      aspectRatios: ['16:9', '9:16'],
+      resolutions: ['720p', '1080p', '4k'],
+      durationRangeSec: { min: 4, max: 8 },
+      seed: true,
+      multipleVideos: true,
+    },
+  },
   {
     id: 'grok-imagine-video',
     name: 'Grok Imagine Video',