wip: zen

3 files changed, 45 insertions, 15 deletions

diff --git a/cloud/app/src/routes/zen/v1/chat/completions.ts b/cloud/app/src/routes/zen/v1/chat/completions.ts
index dc69bb514..13a75a4a4 100644
--- a/cloud/app/src/routes/zen/v1/chat/completions.ts
+++ b/cloud/app/src/routes/zen/v1/chat/completions.ts
@@ -1,8 +1,26 @@
 import type { APIEvent } from "@solidjs/start/server"
 import { handler } from "~/util/zen"
 
+type Usage = {
+  prompt_tokens?: number
+  completion_tokens?: number
+  total_tokens?: number
+  prompt_tokens_details?: {
+    text_tokens?: number
+    audio_tokens?: number
+    image_tokens?: number
+    cached_tokens?: number
+  }
+  completion_tokens_details?: {
+    reasoning_tokens?: number
+    audio_tokens?: number
+    accepted_prediction_tokens?: number
+    rejected_prediction_tokens?: number
+  }
+}
+
 export function POST(input: APIEvent) {
-  let usage: any
+  let usage: Usage
   return handler(input, {
     modifyBody: (body: any) => ({
       ...body,
@@ -17,7 +35,7 @@ export function POST(input: APIEvent) {
 
       let json
       try {
-        json = JSON.parse(chunk.slice(6))
+        json = JSON.parse(chunk.slice(6)) as { usage?: Usage }
       } catch (e) {
         return
       }
@@ -26,11 +44,11 @@ export function POST(input: APIEvent) {
       usage = json.usage
     },
     getStreamUsage: () => usage,
-    normalizeUsage: (usage: any) => ({
+    normalizeUsage: (usage: Usage) => ({
       inputTokens: usage.prompt_tokens ?? 0,
       outputTokens: usage.completion_tokens ?? 0,
-      reasoningTokens: usage.completion_tokens_details?.reasoning_tokens ?? 0,
-      cacheReadTokens: usage.prompt_tokens_details?.cached_tokens ?? 0,
+      reasoningTokens: usage.completion_tokens_details?.reasoning_tokens ?? undefined,
+      cacheReadTokens: usage.prompt_tokens_details?.cached_tokens ?? undefined,
     }),
   })
 }
diff --git a/cloud/app/src/routes/zen/v1/messages.ts b/cloud/app/src/routes/zen/v1/messages.ts
index ff399b034..b2e9e275a 100644
--- a/cloud/app/src/routes/zen/v1/messages.ts
+++ b/cloud/app/src/routes/zen/v1/messages.ts
@@ -53,9 +53,9 @@ export function POST(input: APIEvent) {
     normalizeUsage: (usage: Usage) => ({
       inputTokens: usage.input_tokens ?? 0,
       outputTokens: usage.output_tokens ?? 0,
-      cacheReadTokens: usage.cache_read_input_tokens ?? 0,
-      cacheWrite5mTokens: usage.cache_creation?.ephemeral_5m_input_tokens,
-      cacheWrite1hTokens: usage.cache_creation?.ephemeral_1h_input_tokens,
+      cacheReadTokens: usage.cache_read_input_tokens ?? undefined,
+      cacheWrite5mTokens: usage.cache_creation?.ephemeral_5m_input_tokens ?? undefined,
+      cacheWrite1hTokens: usage.cache_creation?.ephemeral_1h_input_tokens ?? undefined,
     }),
   })
 }
diff --git a/cloud/app/src/routes/zen/v1/responses.ts b/cloud/app/src/routes/zen/v1/responses.ts
index 518a431ab..1bca91f52 100644
--- a/cloud/app/src/routes/zen/v1/responses.ts
+++ b/cloud/app/src/routes/zen/v1/responses.ts
@@ -1,8 +1,20 @@
 import type { APIEvent } from "@solidjs/start/server"
 import { handler } from "~/util/zen"
 
+type Usage = {
+  input_tokens?: number
+  input_tokens_details?: {
+    cached_tokens?: number
+  }
+  output_tokens?: number
+  output_tokens_details?: {
+    reasoning_tokens?: number
+  }
+  total_tokens?: number
+}
+
 export function POST(input: APIEvent) {
-  let usage: any
+  let usage: Usage
   return handler(input, {
     setAuthHeader: (headers: Headers, apiKey: string) => {
       headers.set("authorization", `Bearer ${apiKey}`)
@@ -15,7 +27,7 @@ export function POST(input: APIEvent) {
 
       let json
       try {
-        json = JSON.parse(data.slice(6))
+        json = JSON.parse(data.slice(6)) as { response?: { usage?: Usage } }
       } catch (e) {
         return
       }
@@ -24,14 +36,14 @@ export function POST(input: APIEvent) {
       usage = json.response.usage
     },
     getStreamUsage: () => usage,
-    normalizeUsage: (usage: any) => {
+    normalizeUsage: (usage: Usage) => {
       const inputTokens = usage.input_tokens ?? 0
       const outputTokens = usage.output_tokens ?? 0
-      const reasoningTokens = usage.output_tokens_details?.reasoning_tokens ?? 0
-      const cacheReadTokens = usage.input_tokens_details?.cached_tokens ?? 0
+      const reasoningTokens = usage.output_tokens_details?.reasoning_tokens ?? undefined
+      const cacheReadTokens = usage.input_tokens_details?.cached_tokens ?? undefined
       return {
-        inputTokens: inputTokens - cacheReadTokens,
-        outputTokens: outputTokens - reasoningTokens,
+        inputTokens: inputTokens - (cacheReadTokens ?? 0),
+        outputTokens: outputTokens - (reasoningTokens ?? 0),
         reasoningTokens,
         cacheReadTokens,
       }