[pull] main from CodebuffAI:main

cli/src/components/waiting-room-screen.tsx

@@ -90,6 +90,7 @@ export const WaitingRoomScreen: React.FC<WaitingRoomScreenProps> = ({
     forceStart: true,
     provider: 'gravity',
     fallbackProvider: 'carbon',
+    surface: 'waiting_room',
   })
 
   useFreebuffCtrlCExit()

cli/src/hooks/use-gravity-ad.ts

@@ -35,6 +35,7 @@ export type AdVariant = 'banner' | 'choice'
  * same normalized response shape, so the rest of the hook is provider-agnostic.
  */
 export type AdProvider = 'gravity' | 'carbon'
+export type AdSurface = 'waiting_room'
 
 export type AdData =
   | { variant: 'banner'; ad: AdResponse }
@@ -112,11 +113,14 @@ export const useGravityAd = (options?: {
   provider?: AdProvider
   /** Backup ad network to try when the primary returns no fill or errors. */
   fallbackProvider?: AdProvider
+  /** Product surface requesting the ad. The server maps this to placements. */
+  surface?: AdSurface
 }): GravityAdState => {
   const enabled = options?.enabled ?? true
   const forceStart = options?.forceStart ?? false
   const provider: AdProvider = options?.provider ?? 'gravity'
   const fallbackProvider = options?.fallbackProvider
+  const surface = options?.surface
   const [ad, setAd] = useState<AdResponse | null>(null)
   const [adData, setAdData] = useState<AdData | null>(null)
   const [isLoading, setIsLoading] = useState(false)
@@ -299,6 +303,7 @@ export const useGravityAd = (options?: {
             messages: adMessages,
             sessionId: useChatStore.getState().chatSessionId,
             device: getDeviceInfo(),
+            ...(surface ? { surface } : {}),
             // Carbon requires a real browser-ish useragent for targeting/fraud
             // detection. Gravity ignores it. We source one centrally so every
             // provider that needs it sees the same value.
@@ -430,7 +435,7 @@ export const useGravityAd = (options?: {
       clearInterval(id)
       ctrlRef.current.intervalId = null
     }
-  }, [shouldStart, shouldHideAds, provider, fallbackProvider])
+  }, [shouldStart, shouldHideAds, provider, fallbackProvider, surface])
 
   // Don't return ad when ads should be hidden
   const visible = shouldStart && !shouldHideAds

scripts/test-canopywave-long.ts

@@ -33,11 +33,10 @@ const MODEL_CONFIGS: Record<string, ModelConfig> = {
     outputCostPerToken: 1.20 / 1_000_000,
   },
   kimi: {
-    // Pricing is approximate — based on public Moonshot k2 rates; CanopyWave may differ.
     id: 'moonshotai/kimi-k2.6',
-    inputCostPerToken: 0.60 / 1_000_000,
-    cachedInputCostPerToken: 0.15 / 1_000_000,
-    outputCostPerToken: 2.50 / 1_000_000,
+    inputCostPerToken: 0.95 / 1_000_000,
+    cachedInputCostPerToken: 0.16 / 1_000_000,
+    outputCostPerToken: 4.00 / 1_000_000,
   },
 }
 

web/src/app/api/v1/ads/_post.ts

@@ -35,12 +35,14 @@ const deviceSchema = z.object({
 })
 
 const providerSchema = z.enum(['gravity', 'carbon']).default('gravity')
+const surfaceSchema = z.enum(['waiting_room'])
 
 const bodySchema = z.object({
   provider: providerSchema.optional(),
   messages: z.array(messageSchema).optional().default([]),
   sessionId: z.string().optional(),
   device: deviceSchema.optional(),
+  surface: surfaceSchema.optional(),
   /** Browser/CLI useragent passed through to providers that require it. */
   userAgent: z.string().optional(),
 })
@@ -136,6 +138,7 @@ export async function postAds(params: {
       clientIp,
       userAgent,
       device: parsedBody.device,
+      surface: parsedBody.surface,
       messages: parsedBody.messages,
       testMode: serverEnv.CB_ENVIRONMENT !== 'prod',
       logger,

web/src/app/api/v1/chat/completions/_post.ts

@@ -532,9 +532,10 @@ export async function postChatCompletions(params: {
       if (bodyStream) {
         // Streaming request — route to SiliconFlow/CanopyWave/Fireworks for supported models
         const useSiliconFlow = false // isSiliconFlowModel(typedBody.model)
-        const useCanopyWave = false // isCanopyWaveModel(typedBody.model)
-        const useFireworks = isFireworksModel(typedBody.model)
-        const useOpenAIDirect = !useFireworks && isOpenAIDirectModel(typedBody.model)
+        const useCanopyWave = isCanopyWaveModel(typedBody.model)
+        const useFireworks = !useCanopyWave && isFireworksModel(typedBody.model)
+        const useOpenAIDirect =
+          !useCanopyWave && !useFireworks && isOpenAIDirectModel(typedBody.model)
         const stream = useSiliconFlow
           ? await handleSiliconFlowStream({
             body: typedBody,
@@ -606,12 +607,12 @@ export async function postChatCompletions(params: {
         })
       } else {
         // Non-streaming request — route to SiliconFlow/CanopyWave/Fireworks for supported models
-        // TEMPORARILY DISABLED: route through OpenRouter
         const model = typedBody.model
         const useSiliconFlow = false // isSiliconFlowModel(model)
-        const useCanopyWave = false // isCanopyWaveModel(model)
-        const useFireworks = isFireworksModel(model)
-        const shouldUseOpenAIEndpoint = !useFireworks && isOpenAIDirectModel(model)
+        const useCanopyWave = isCanopyWaveModel(model)
+        const useFireworks = !useCanopyWave && isFireworksModel(model)
+        const shouldUseOpenAIEndpoint =
+          !useCanopyWave && !useFireworks && isOpenAIDirectModel(model)
 
         const nonStreamRequest = useSiliconFlow
           ? handleSiliconFlowNonStream({

web/src/lib/ad-providers/gravity.ts

@@ -19,6 +19,12 @@ const CHOICE_PLACEMENT_IDS = [
   'choice-ad-3',
   'choice-ad-4',
 ]
+const WAITING_ROOM_PLACEMENT_IDS = [
+  'waiting-room-1',
+  'waiting-room-2',
+  'waiting-room-3',
+  'waiting-room-4',
+]
 
 type GravityRawAd = {
   adText: string
@@ -105,16 +111,21 @@ export function createGravityProvider(config: { apiKey: string }): AdProvider {
         fetch,
       } = input
 
-      const variant = getGravityVariant(userId)
+      const variant =
+        input.surface === 'waiting_room' ? 'choice' : getGravityVariant(userId)
       const filteredMessages = prepareGravityMessages(messages)
 
-      const placements =
-        variant === 'choice'
-          ? CHOICE_PLACEMENT_IDS.map((id) => ({
-              placement: 'below_response',
-              placement_id: id,
-            }))
-          : [{ placement: 'below_response', placement_id: BANNER_PLACEMENT_ID }]
+      const placementIds =
+        input.surface === 'waiting_room'
+          ? WAITING_ROOM_PLACEMENT_IDS
+          : variant === 'choice'
+          ? CHOICE_PLACEMENT_IDS
+          : [BANNER_PLACEMENT_ID]
+
+      const placements = placementIds.map((id) => ({
+        placement: 'below_response',
+        placement_id: id,
+      }))
 
       const deviceBody = clientIp
         ? {

web/src/lib/ad-providers/types.ts

@@ -41,6 +41,8 @@ export type AdDeviceInfo = {
   locale?: string
 }
 
+export type AdSurface = 'waiting_room'
+
 export type FetchAdInput = {
   userId: string
   userEmail: string | null
@@ -50,6 +52,8 @@ export type FetchAdInput = {
   /** Browser/CLI useragent string, passed through to upstream. */
   userAgent?: string
   device?: AdDeviceInfo
+  /** Product surface requesting the ad. Providers may map this to placements. */
+  surface?: AdSurface
   /** Last user + last preceding assistant message, if any. Used by Gravity. */
   messages?: AdMessage[]
   /** Set in non-prod so providers can request test ads. */

web/src/llm-api/canopywave.ts

@@ -26,17 +26,52 @@ const canopywaveAgent = new Agent({
   bodyTimeout: 0,
 })
 
-/** Map from OpenRouter model IDs to CanopyWave model IDs */
-const CANOPYWAVE_MODEL_MAP: Record<string, string> = {
-  'minimax/minimax-m2.5': 'minimax/minimax-m2.5',
+// CanopyWave per-token pricing (dollars per token)
+interface CanopyWavePricing {
+  inputCostPerToken: number
+  cachedInputCostPerToken: number
+  outputCostPerToken: number
+}
+
+/** Single source of truth: which OpenRouter model IDs we route through
+ *  CanopyWave, the corresponding CanopyWave model ID, and per-model pricing.
+ *  Kept as one map so adding a model can't drift between routing and billing. */
+const CANOPYWAVE_MODELS: Record<
+  string,
+  { canopywaveId: string; pricing: CanopyWavePricing }
+> = {
+  'minimax/minimax-m2.5': {
+    canopywaveId: 'minimax/minimax-m2.5',
+    pricing: {
+      inputCostPerToken: 0.27 / 1_000_000,
+      cachedInputCostPerToken: 0.03 / 1_000_000,
+      outputCostPerToken: 1.08 / 1_000_000,
+    },
+  },
+  'moonshotai/kimi-k2.6': {
+    canopywaveId: 'moonshotai/kimi-k2.6',
+    pricing: {
+      inputCostPerToken: 0.95 / 1_000_000,
+      cachedInputCostPerToken: 0.16 / 1_000_000,
+      outputCostPerToken: 4.00 / 1_000_000,
+    },
+  },
 }
 
 export function isCanopyWaveModel(model: string): boolean {
-  return model in CANOPYWAVE_MODEL_MAP
+  return model in CANOPYWAVE_MODELS
 }
 
 function getCanopyWaveModelId(openrouterModel: string): string {
-  return CANOPYWAVE_MODEL_MAP[openrouterModel] ?? openrouterModel
+  return CANOPYWAVE_MODELS[openrouterModel]?.canopywaveId ?? openrouterModel
+}
+
+function getCanopyWavePricing(model: string): CanopyWavePricing {
+  const entry = CANOPYWAVE_MODELS[model]
+  if (!entry) {
+    throw new Error(`No CanopyWave pricing found for model: ${model}`)
+  }
+  return entry.pricing
 }
 
 type StreamState = { responseText: string; reasoningText: string; ttftMs: number | null; billedAlready: boolean }
@@ -85,12 +120,7 @@ function createCanopyWaveRequest(params: {
   })
 }
 
-// CanopyWave per-token pricing (dollars per token) for MiniMax M2.5
-const CANOPYWAVE_INPUT_COST_PER_TOKEN = 0.27 / 1_000_000
-const CANOPYWAVE_CACHED_INPUT_COST_PER_TOKEN = 0.03 / 1_000_000
-const CANOPYWAVE_OUTPUT_COST_PER_TOKEN = 1.08 / 1_000_000
-
-function extractUsageAndCost(usage: Record<string, unknown> | undefined | null): UsageData {
+function extractUsageAndCost(usage: Record<string, unknown> | undefined | null, model: string): UsageData {
   if (!usage) return { inputTokens: 0, outputTokens: 0, cacheReadInputTokens: 0, reasoningTokens: 0, cost: 0 }
   const promptDetails = usage.prompt_tokens_details as Record<string, unknown> | undefined | null
   const completionDetails = usage.completion_tokens_details as Record<string, unknown> | undefined | null
@@ -100,11 +130,12 @@ function extractUsageAndCost(usage: Record<string, unknown> | undefined | null):
   const cacheReadInputTokens = typeof promptDetails?.cached_tokens === 'number' ? promptDetails.cached_tokens : 0
   const reasoningTokens = typeof completionDetails?.reasoning_tokens === 'number' ? completionDetails.reasoning_tokens : 0
 
+  const pricing = getCanopyWavePricing(model)
   const nonCachedInputTokens = Math.max(0, inputTokens - cacheReadInputTokens)
   const cost =
-    nonCachedInputTokens * CANOPYWAVE_INPUT_COST_PER_TOKEN +
-    cacheReadInputTokens * CANOPYWAVE_CACHED_INPUT_COST_PER_TOKEN +
-    outputTokens * CANOPYWAVE_OUTPUT_COST_PER_TOKEN
+    nonCachedInputTokens * pricing.inputCostPerToken +
+    cacheReadInputTokens * pricing.cachedInputCostPerToken +
+    outputTokens * pricing.outputCostPerToken
 
   return { inputTokens, outputTokens, cacheReadInputTokens, reasoningTokens, cost }
 }
@@ -139,7 +170,7 @@ export async function handleCanopyWaveNonStream({
   const data = await response.json()
   const content = data.choices?.[0]?.message?.content ?? ''
   const reasoningText = data.choices?.[0]?.message?.reasoning_content ?? data.choices?.[0]?.message?.reasoning ?? ''
-  const usageData = extractUsageAndCost(data.usage)
+  const usageData = extractUsageAndCost(data.usage, originalModel)
 
   insertMessageToBigQuery({
     messageId: data.id,
@@ -453,7 +484,7 @@ async function handleResponse({
     return { state }
   }
 
-  const usageData = extractUsageAndCost(data.usage as Record<string, unknown>)
+  const usageData = extractUsageAndCost(data.usage as Record<string, unknown>, originalModel)
   const messageId = typeof data.id === 'string' ? data.id : 'unknown'
 
   state.billedAlready = true