fix(provider): select prompt cache keys by SDK

packages/opencode/src/provider/transform.ts

@@ -58,6 +58,28 @@ function sdkKey(npm: string): string | undefined {
       return "vertex"
     case "@ai-sdk/google":
       return "google"
+    case "@ai-sdk/alibaba":
+      return "alibaba"
+    case "@ai-sdk/cerebras":
+      return "cerebras"
+    case "@ai-sdk/cohere":
+      return "cohere"
+    case "@ai-sdk/deepinfra":
+      return "deepinfra"
+    case "@ai-sdk/groq":
+      return "groq"
+    case "@ai-sdk/mistral":
+      return "mistral"
+    case "@ai-sdk/perplexity":
+      return "perplexity"
+    case "@ai-sdk/togetherai":
+      return "togetherai"
+    case "@ai-sdk/vercel":
+      return "vercel"
+    case "@ai-sdk/xai":
+      return "xai"
+    case "venice-ai-sdk-provider":
+      return "venice"
     case "@ai-sdk/gateway":
       return "gateway"
     case "@openrouter/ai-sdk-provider":
@@ -442,6 +464,9 @@ function mapProviderOptions(
 export function message(msgs: ModelMessage[], model: Provider.Model, options: Record<string, unknown>) {
   msgs = unsupportedParts(msgs, model)
   msgs = normalizeMessages(msgs, model, options)
+  const usesAnthropicAutomaticCaching =
+    options.cacheControl !== undefined &&
+    (model.api.npm === "@ai-sdk/anthropic" || model.api.npm === "@ai-sdk/google-vertex/anthropic")
   if (
     (model.providerID === "anthropic" ||
       model.providerID === "google-vertex-anthropic" ||
@@ -451,7 +476,8 @@ export function message(msgs: ModelMessage[], model: Provider.Model, options: Re
       model.id.includes("claude") ||
       model.api.npm === "@ai-sdk/anthropic" ||
       model.api.npm === "@ai-sdk/alibaba") &&
-    model.api.npm !== "@ai-sdk/gateway"
+    model.api.npm !== "@ai-sdk/gateway" &&
+    !usesAnthropicAutomaticCaching
   ) {
     msgs = applyCaching(msgs, model)
   }
@@ -1137,7 +1163,6 @@ export function options(input: {
 
   if (input.model.api.npm === "@ai-sdk/azure") {
     result["store"] = false
-    result["promptCacheKey"] = input.sessionID
   }
 
   if (input.model.api.npm === "@openrouter/ai-sdk-provider" || input.model.api.npm === "@llmgateway/ai-sdk-provider") {
@@ -1166,16 +1191,6 @@ export function options(input: {
     }
   }
 
-  if (
-    input.providerOptions?.setCacheKey !== false &&
-    (input.model.providerID === "openai" ||
-      input.model.api.npm === "@ai-sdk/openai" ||
-      input.model.api.npm === "@ai-sdk/xai" ||
-      input.providerOptions?.setCacheKey)
-  ) {
-    result["promptCacheKey"] = input.sessionID
-  }
-
   if (input.model.providerID === "meta" && input.model.api.npm === "@ai-sdk/openai") {
     result["reasoningSummary"] = "auto"
     result["include"] = INCLUDE_ENCRYPTED_REASONING
@@ -1224,6 +1239,25 @@ export function options(input: {
     result["enable_thinking"] = true
   }
 
+  if (input.providerOptions?.setCacheKey !== false) {
+    if (input.model.api.npm === "@ai-sdk/deepinfra" || input.model.api.npm === "@ai-sdk/cerebras") {
+      result["prompt_cache_key"] = input.sessionID
+    } else if (
+      input.model.api.npm === "@ai-sdk/openai" ||
+      input.model.api.npm === "@ai-sdk/azure" ||
+      input.model.api.npm === "@ai-sdk/xai" ||
+      input.model.api.npm === "@ai-sdk/mistral" ||
+      input.model.api.npm === "venice-ai-sdk-provider" ||
+      input.providerOptions?.setCacheKey === true
+    ) {
+      result["promptCacheKey"] = input.sessionID
+    }
+  }
+
+  if (input.model.api.npm === "@ai-sdk/gateway") {
+    result["gateway"] = { caching: "auto" }
+  }
+
   if (input.model.api.npm === "@ai-sdk/azure" && input.model.api.id.includes("gpt-5.5")) {
     result["reasoningSummary"] = "auto"
     return result
@@ -1256,26 +1290,13 @@ export function options(input: {
       result["textVerbosity"] = "low"
     }
 
-    if (input.model.providerID.startsWith("opencode")) {
+    if (input.model.providerID.startsWith("opencode") && input.providerOptions?.setCacheKey !== false) {
       result["promptCacheKey"] = input.sessionID
       result["include"] = INCLUDE_ENCRYPTED_REASONING
       result["reasoningSummary"] = "auto"
     }
   }
 
-  if (input.model.providerID === "venice") {
-    result["promptCacheKey"] = input.sessionID
-  }
-
-  if (input.model.providerID === "openrouter") {
-    result["prompt_cache_key"] = input.sessionID
-  }
-  if (input.model.api.npm === "@ai-sdk/gateway") {
-    result["gateway"] = {
-      caching: "auto",
-    }
-  }
-
   return result
 }
 

packages/opencode/test/provider/transform.test.ts

@@ -88,6 +88,32 @@ describe("ProviderTransform.options - setCacheKey", () => {
     expect(result.promptCacheKey).toBe(sessionID)
   })
 
+  test("should set promptCacheKey for the OpenAI SDK regardless of provider ID", () => {
+    const result = ProviderTransform.options({
+      model: {
+        ...mockModel,
+        providerID: "custom-openai",
+        api: { id: "gpt-5", url: "https://example.com", npm: "@ai-sdk/openai" },
+      },
+      sessionID,
+      providerOptions: {},
+    })
+    expect(result.promptCacheKey).toBe(sessionID)
+  })
+
+  test("should not set promptCacheKey for the OpenAI-compatible SDK by provider name", () => {
+    const result = ProviderTransform.options({
+      model: {
+        ...mockModel,
+        providerID: "openai",
+        api: { id: "gpt-5", url: "https://example.com", npm: "@ai-sdk/openai-compatible" },
+      },
+      sessionID,
+      providerOptions: {},
+    })
+    expect(result.promptCacheKey).toBeUndefined()
+  })
+
   test("should not set promptCacheKey for openai when explicitly disabled", () => {
     const openaiModel = {
       ...mockModel,
@@ -209,6 +235,70 @@ describe("ProviderTransform.options - setCacheKey", () => {
       providerOptions: {},
     })
     expect(result.store).toBe(false)
+    expect(result.promptCacheKey).toBe(sessionID)
+  })
+
+  test("should disable the Azure cache key without disabling store=false", () => {
+    const result = ProviderTransform.options({
+      model: {
+        ...mockModel,
+        providerID: "azure",
+        api: { id: "gpt-5", url: "https://azure.com", npm: "@ai-sdk/azure" },
+      },
+      sessionID,
+      providerOptions: { setCacheKey: false },
+    })
+    expect(result.store).toBe(false)
+    expect(result.promptCacheKey).toBeUndefined()
+  })
+
+  test("should keep the Azure cache key for gpt-5.5 early return", () => {
+    const result = ProviderTransform.options({
+      model: {
+        ...mockModel,
+        providerID: "azure",
+        api: { id: "gpt-5.5", url: "https://azure.com", npm: "@ai-sdk/azure" },
+      },
+      sessionID,
+      providerOptions: {},
+    })
+    expect(result.store).toBe(false)
+    expect(result.reasoningSummary).toBe("auto")
+    expect(result.promptCacheKey).toBe(sessionID)
+  })
+
+  for (const npm of ["@ai-sdk/deepinfra", "@ai-sdk/cerebras"]) {
+    test(`should set the snake-case cache key for ${npm}`, () => {
+      const result = ProviderTransform.options({
+        model: { ...mockModel, providerID: "custom", api: { ...mockModel.api, npm } },
+        sessionID,
+        providerOptions: {},
+      })
+      expect(result.prompt_cache_key).toBe(sessionID)
+      expect(result.promptCacheKey).toBeUndefined()
+    })
+  }
+
+  test("should set promptCacheKey for the Mistral SDK", () => {
+    const result = ProviderTransform.options({
+      model: { ...mockModel, providerID: "custom", api: { ...mockModel.api, npm: "@ai-sdk/mistral" } },
+      sessionID,
+      providerOptions: {},
+    })
+    expect(result.promptCacheKey).toBe(sessionID)
+  })
+
+  test("should not send an undocumented OpenRouter prompt_cache_key", () => {
+    const result = ProviderTransform.options({
+      model: {
+        ...mockModel,
+        providerID: "openrouter",
+        api: { ...mockModel.api, npm: "@openrouter/ai-sdk-provider" },
+      },
+      sessionID,
+      providerOptions: {},
+    })
+    expect(result.prompt_cache_key).toBeUndefined()
   })
 })
 
@@ -722,6 +812,17 @@ describe("ProviderTransform.providerOptions", () => {
     })
   })
 
+  test("uses canonical sdk key for custom xAI models", () => {
+    const model = createModel({
+      providerID: "my-xai",
+      api: { id: "grok-4", url: "https://api.x.ai", npm: "@ai-sdk/xai" },
+    })
+
+    expect(ProviderTransform.providerOptions(model, { promptCacheKey: "session" })).toEqual({
+      xai: { promptCacheKey: "session" },
+    })
+  })
+
   test("forces reasoning for explicit effort even when model is not marked reasoning-capable", () => {
     const model = createModel({
       capabilities: {
@@ -3011,6 +3112,20 @@ describe("ProviderTransform.message - cache control on gateway", () => {
     })
   })
 
+  test("does not add explicit breakpoints when Anthropic automatic caching is enabled", () => {
+    const model = createModel({
+      providerID: "anthropic",
+      api: { id: "claude-sonnet-4", url: "https://api.anthropic.com", npm: "@ai-sdk/anthropic" },
+    })
+    const msgs = [
+      { role: "system", content: "You are a helpful assistant" },
+      { role: "user", content: "Hello" },
+    ] as any[]
+
+    const result = ProviderTransform.message(msgs, model, { cacheControl: { type: "ephemeral" } }) as any[]
+    expect(result.every((message) => message.providerOptions === undefined)).toBe(true)
+  })
+
   test("google-vertex-anthropic applies cache control", () => {
     const model = createModel({
       providerID: "google-vertex-anthropic",