Feat: Add support for AWS Bedrock Anthropic prompt caching (#2034)

* Add Bedrock prompt caching support (optional)

This feature protected under checkbox because it is not yet rolled out
to everyone, and if you will try to send cache headers, and its not
enabled for you, you will get error

* Add changeset

* Update supported models

* Fix copy

---------

Co-authored-by: Saoud Rizwan <7799382+saoudrizwan@users.noreply.github.com>

.changeset/.mighty-tools-remain.md

@@ -0,0 +1,7 @@
+---
+"claude-dev": minor
+---
+Add Bedrock prompt caching support (optional).
+    
+This feature protected under checkbox because it is not yet rolled out to everyone, and if you will try to send cache headers, and its not enabled for you, you will get error.
+

src/api/providers/bedrock.ts

@@ -19,9 +19,15 @@ export class AwsBedrockHandler implements ApiHandler {
 		// cross region inference requires prefixing the model id with the region
 		let modelId = await this.getModelId()
 
-		// create anthropic client, using sessions created or renewed after this handler's
+		// Get model info and message indices for caching
+		const model = this.getModel()
+		const userMsgIndices = messages.reduce((acc, msg, index) => (msg.role === "user" ? [...acc, index] : acc), [] as number[])
+		const lastUserMsgIndex = userMsgIndices[userMsgIndices.length - 1] ?? -1
+		const secondLastMsgUserIndex = userMsgIndices[userMsgIndices.length - 2] ?? -1
+
+		// Create anthropic client, using sessions created or renewed after this handler's
 		// initialization, and allowing for session renewal if necessary as well
-		let client = await this.getClient()
+		const client = await this.getClient()
 
 		const stream = await client.messages.create({
 			model: modelId,
@@ -32,6 +38,7 @@ export class AwsBedrockHandler implements ApiHandler {
 			messages,
 			stream: true,
 		})
+
 		for await (const chunk of stream) {
 			switch (chunk.type) {
 				case "message_start":
@@ -40,6 +47,8 @@ export class AwsBedrockHandler implements ApiHandler {
 						type: "usage",
 						inputTokens: usage.input_tokens || 0,
 						outputTokens: usage.output_tokens || 0,
+						cacheWriteTokens: usage.cache_creation_input_tokens || undefined,
+						cacheReadTokens: usage.cache_read_input_tokens || undefined,
 					}
 					break
 				case "message_delta":
@@ -49,7 +58,6 @@ export class AwsBedrockHandler implements ApiHandler {
 						outputTokens: chunk.usage.output_tokens || 0,
 					}
 					break
-
 				case "content_block_start":
 					switch (chunk.content_block.type) {
 						case "text":
@@ -129,11 +137,9 @@ export class AwsBedrockHandler implements ApiHandler {
 					return `us.${this.getModel().id}`
 				case "eu-":
 					return `eu.${this.getModel().id}`
-					break
 				default:
 					// cross region inference is not supported in this region, falling back to default model
 					return this.getModel().id
-					break
 			}
 		}
 		return this.getModel().id
@@ -150,7 +156,7 @@ export class AwsBedrockHandler implements ApiHandler {
 		}
 	}
 
-	private static async setEnv(key: string, value: string | undefined) {
+	private static setEnv(key: string, value: string | undefined) {
 		if (key !== "" && value !== undefined) {
 			process.env[key] = value
 		}

src/core/webview/ClineProvider.ts

@@ -65,6 +65,7 @@ type GlobalStateKey =
 	| "apiModelId"
 	| "awsRegion"
 	| "awsUseCrossRegionInference"
+	| "awsBedrockUsePromptCache"
 	| "awsProfile"
 	| "awsUseProfile"
 	| "vertexProjectId"
@@ -565,6 +566,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 								awsSessionToken,
 								awsRegion,
 								awsUseCrossRegionInference,
+								awsBedrockUsePromptCache,
 								awsProfile,
 								awsUseProfile,
 								vertexProjectId,
@@ -607,6 +609,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 							await this.storeSecret("awsSessionToken", awsSessionToken)
 							await this.updateGlobalState("awsRegion", awsRegion)
 							await this.updateGlobalState("awsUseCrossRegionInference", awsUseCrossRegionInference)
+							await this.updateGlobalState("awsBedrockUsePromptCache", awsBedrockUsePromptCache)
 							await this.updateGlobalState("awsProfile", awsProfile)
 							await this.updateGlobalState("awsUseProfile", awsUseProfile)
 							await this.updateGlobalState("vertexProjectId", vertexProjectId)
@@ -1848,6 +1851,7 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 			awsSessionToken,
 			awsRegion,
 			awsUseCrossRegionInference,
+			awsBedrockUsePromptCache,
 			awsProfile,
 			awsUseProfile,
 			vertexProjectId,
@@ -1903,6 +1907,7 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 			this.getSecret("awsSessionToken") as Promise<string | undefined>,
 			this.getGlobalState("awsRegion") as Promise<string | undefined>,
 			this.getGlobalState("awsUseCrossRegionInference") as Promise<boolean | undefined>,
+			this.getGlobalState("awsBedrockUsePromptCache") as Promise<boolean | undefined>,
 			this.getGlobalState("awsProfile") as Promise<string | undefined>,
 			this.getGlobalState("awsUseProfile") as Promise<boolean | undefined>,
 			this.getGlobalState("vertexProjectId") as Promise<string | undefined>,
@@ -1981,6 +1986,7 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 				awsSessionToken,
 				awsRegion,
 				awsUseCrossRegionInference,
+				awsBedrockUsePromptCache,
 				awsProfile,
 				awsUseProfile,
 				vertexProjectId,

src/shared/api.ts

@@ -32,6 +32,7 @@ export interface ApiHandlerOptions {
 	awsSessionToken?: string
 	awsRegion?: string
 	awsUseCrossRegionInference?: boolean
+	awsBedrockUsePromptCache?: boolean
 	awsUseProfile?: boolean
 	awsProfile?: string
 	vertexProjectId?: string
@@ -160,17 +161,21 @@ export const bedrockModels = {
 		contextWindow: 200_000,
 		supportsImages: true,
 		supportsComputerUse: true,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 3.0,
 		outputPrice: 15.0,
+		cacheWritesPrice: 3.75,
+		cacheReadsPrice: 0.3,
 	},
 	"anthropic.claude-3-5-haiku-20241022-v1:0": {
 		maxTokens: 8192,
 		contextWindow: 200_000,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 1.0,
 		outputPrice: 5.0,
+		cacheWritesPrice: 1.0,
+		cacheReadsPrice: 0.08,
 	},
 	"anthropic.claude-3-5-sonnet-20240620-v1:0": {
 		maxTokens: 8192,

webview-ui/src/components/settings/ApiOptions.tsx

@@ -537,17 +537,35 @@ const ApiOptions = ({ showModelOptions, apiErrorMessage, modelIdErrorMessage, is
 							{/* <VSCodeOption value="us-gov-east-1">us-gov-east-1</VSCodeOption> */}
 						</VSCodeDropdown>
 					</DropdownContainer>
-					<VSCodeCheckbox
-						checked={apiConfiguration?.awsUseCrossRegionInference || false}
-						onChange={(e: any) => {
-							const isChecked = e.target.checked === true
-							setApiConfiguration({
-								...apiConfiguration,
-								awsUseCrossRegionInference: isChecked,
-							})
-						}}>
-						Use cross-region inference
-					</VSCodeCheckbox>
+					<div style={{ display: "flex", flexDirection: "column" }}>
+						<VSCodeCheckbox
+							checked={apiConfiguration?.awsUseCrossRegionInference || false}
+							onChange={(e: any) => {
+								const isChecked = e.target.checked === true
+								setApiConfiguration({
+									...apiConfiguration,
+									awsUseCrossRegionInference: isChecked,
+								})
+							}}>
+							Use cross-region inference
+						</VSCodeCheckbox>
+
+						{selectedModelInfo.supportsPromptCache && (
+							<>
+								<VSCodeCheckbox
+									checked={apiConfiguration?.awsBedrockUsePromptCache || false}
+									onChange={(e: any) => {
+										const isChecked = e.target.checked === true
+										setApiConfiguration({
+											...apiConfiguration,
+											awsBedrockUsePromptCache: isChecked,
+										})
+									}}>
+									Use prompt caching (Beta)
+								</VSCodeCheckbox>
+							</>
+						)}
+					</div>
 					<p
 						style={{
 							fontSize: "12px",