feat: add prompt_cache_key

src/api/index.ts

@@ -44,7 +44,7 @@ import {
 import { NativeOllamaHandler } from "./providers/native-ollama"
 
 export interface SingleCompletionHandler {
-	completePrompt(prompt: string): Promise<string>
+	completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string>
 }
 
 export interface ApiHandlerCreateMessageMetadata {
@@ -65,6 +65,7 @@ export interface ApiHandlerCreateMessageMetadata {
 	 * @default true
 	 */
 	store?: boolean
+	safetyIdentifier?: string
 }
 
 export interface ApiHandler {

src/api/providers/anthropic-vertex.ts

@@ -175,7 +175,7 @@ export class AnthropicVertexHandler extends BaseProvider implements SingleComple
 		return { id: id.endsWith(":thinking") ? id.replace(":thinking", "") : id, info, ...params }
 	}
 
-	async completePrompt(prompt: string) {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata) {
 		try {
 			let {
 				id,

src/api/providers/anthropic.ts

@@ -278,7 +278,7 @@ export class AnthropicHandler extends BaseProvider implements SingleCompletionHa
 		}
 	}
 
-	async completePrompt(prompt: string) {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata) {
 		let { id: model, temperature } = this.getModel()
 
 		const message = await this.client.messages.create({

src/api/providers/base-openai-compatible-provider.ts

@@ -83,6 +83,8 @@ export abstract class BaseOpenAiCompatibleProvider<ModelName extends string>
 			messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
 			stream: true,
 			stream_options: { include_usage: true },
+			prompt_cache_key: metadata?.taskId,
+			safety_identifier: metadata?.safetyIdentifier,
 		}
 
 		try {
@@ -119,13 +121,15 @@ export abstract class BaseOpenAiCompatibleProvider<ModelName extends string>
 		}
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		const { id: modelId } = this.getModel()
 
 		try {
 			const response = await this.client.chat.completions.create({
 				model: modelId,
 				messages: [{ role: "user", content: prompt }],
+				prompt_cache_key: metadata?.taskId,
+				safety_identifier: metadata?.safetyIdentifier,
 			})
 
 			return response.choices[0]?.message.content || ""

src/api/providers/bedrock.ts

@@ -633,7 +633,7 @@ export class AwsBedrockHandler extends BaseProvider implements SingleCompletionH
 		}
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		try {
 			const modelConfig = this.getModel()
 

src/api/providers/cerebras.ts

@@ -277,14 +277,16 @@ export class CerebrasHandler extends BaseProvider implements SingleCompletionHan
 		}
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		const { id: model } = this.getModel()
 
 		// Prepare request body for non-streaming completion
 		const requestBody = {
 			model,
 			messages: [{ role: "user", content: prompt }],
 			stream: false,
+			prompt_cache_key: metadata?.taskId,
+			safety_identifier: metadata?.safetyIdentifier,
 		}
 
 		try {

src/api/providers/chutes.ts

@@ -9,6 +9,7 @@ import { convertToOpenAiMessages } from "../transform/openai-format"
 import { ApiStream } from "../transform/stream"
 
 import { BaseOpenAiCompatibleProvider } from "./base-openai-compatible-provider"
+import { ApiHandlerCreateMessageMetadata } from ".."
 
 export class ChutesHandler extends BaseOpenAiCompatibleProvider<ChutesModelId> {
 	constructor(options: ApiHandlerOptions) {
@@ -44,13 +45,19 @@ export class ChutesHandler extends BaseOpenAiCompatibleProvider<ChutesModelId> {
 		}
 	}
 
-	override async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
+	override async *createMessage(
+		systemPrompt: string,
+		messages: Anthropic.Messages.MessageParam[],
+		metadata?: ApiHandlerCreateMessageMetadata,
+	): ApiStream {
 		const model = this.getModel()
 
 		if (model.id.includes("DeepSeek-R1")) {
 			const stream = await this.client.chat.completions.create({
 				...this.getCompletionParams(systemPrompt, messages),
 				messages: convertToR1Format([{ role: "user", content: systemPrompt }, ...messages]),
+				prompt_cache_key: metadata?.taskId,
+				safety_identifier: metadata?.safetyIdentifier,
 			})
 
 			const matcher = new XmlMatcher(
@@ -85,7 +92,7 @@ export class ChutesHandler extends BaseOpenAiCompatibleProvider<ChutesModelId> {
 				yield processedChunk
 			}
 		} else {
-			yield* super.createMessage(systemPrompt, messages)
+			yield* super.createMessage(systemPrompt, messages, metadata)
 		}
 	}
 

src/api/providers/deepinfra.ts

@@ -60,18 +60,15 @@ export class DeepInfraHandler extends RouterProvider implements SingleCompletion
 		// Ensure we have up-to-date model metadata
 		await this.fetchModel()
 		const { id: modelId, info, reasoningEffort: reasoning_effort } = await this.fetchModel()
-		let prompt_cache_key = undefined
-		if (info.supportsPromptCache && _metadata?.taskId) {
-			prompt_cache_key = _metadata.taskId
-		}
 
 		const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {
 			model: modelId,
 			messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
 			stream: true,
 			stream_options: { include_usage: true },
 			reasoning_effort,
-			prompt_cache_key,
+			prompt_cache_key: _metadata?.taskId,
+			safety_identifier: _metadata?.safetyIdentifier,
 		} as OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming
 
 		if (this.supportsTemperature(modelId)) {
@@ -106,13 +103,15 @@ export class DeepInfraHandler extends RouterProvider implements SingleCompletion
 		}
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		await this.fetchModel()
 		const { id: modelId, info } = this.getModel()
 
 		const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsNonStreaming = {
 			model: modelId,
 			messages: [{ role: "user", content: prompt }],
+			prompt_cache_key: metadata?.taskId,
+			safety_identifier: metadata?.safetyIdentifier,
 		}
 		if (this.supportsTemperature(modelId)) {
 			requestOptions.temperature = this.options.modelTemperature ?? 0

src/api/providers/fake-ai.ts

@@ -28,7 +28,7 @@ interface FakeAI {
 	): ApiStream
 	getModel(): { id: string; info: ModelInfo }
 	countTokens(content: Array<Anthropic.Messages.ContentBlockParam>): Promise<number>
-	completePrompt(prompt: string): Promise<string>
+	completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string>
 }
 
 /**
@@ -75,7 +75,7 @@ export class FakeAIHandler implements ApiHandler, SingleCompletionHandler {
 		return this.ai.countTokens(content)
 	}
 
-	completePrompt(prompt: string): Promise<string> {
+	completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		return this.ai.completePrompt(prompt)
 	}
 }

src/api/providers/featherless.ts

@@ -1,4 +1,9 @@
-import { DEEP_SEEK_DEFAULT_TEMPERATURE, type FeatherlessModelId, featherlessDefaultModelId, featherlessModels } from "@roo-code/types"
+import {
+	DEEP_SEEK_DEFAULT_TEMPERATURE,
+	type FeatherlessModelId,
+	featherlessDefaultModelId,
+	featherlessModels,
+} from "@roo-code/types"
 import { Anthropic } from "@anthropic-ai/sdk"
 import OpenAI from "openai"
 
@@ -9,6 +14,7 @@ import { convertToOpenAiMessages } from "../transform/openai-format"
 import { ApiStream } from "../transform/stream"
 
 import { BaseOpenAiCompatibleProvider } from "./base-openai-compatible-provider"
+import { ApiHandlerCreateMessageMetadata } from ".."
 
 export class FeatherlessHandler extends BaseOpenAiCompatibleProvider<FeatherlessModelId> {
 	constructor(options: ApiHandlerOptions) {
@@ -26,6 +32,7 @@ export class FeatherlessHandler extends BaseOpenAiCompatibleProvider<Featherless
 	private getCompletionParams(
 		systemPrompt: string,
 		messages: Anthropic.Messages.MessageParam[],
+		metadata?: ApiHandlerCreateMessageMetadata,
 	): OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming {
 		const {
 			id: model,
@@ -41,15 +48,21 @@ export class FeatherlessHandler extends BaseOpenAiCompatibleProvider<Featherless
 			messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
 			stream: true,
 			stream_options: { include_usage: true },
+			prompt_cache_key: metadata?.taskId,
+			safety_identifier: metadata?.safetyIdentifier,
 		}
 	}
 
-	override async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
+	override async *createMessage(
+		systemPrompt: string,
+		messages: Anthropic.Messages.MessageParam[],
+		metadata?: ApiHandlerCreateMessageMetadata,
+	): ApiStream {
 		const model = this.getModel()
 
 		if (model.id.includes("DeepSeek-R1")) {
 			const stream = await this.client.chat.completions.create({
-				...this.getCompletionParams(systemPrompt, messages),
+				...this.getCompletionParams(systemPrompt, messages, metadata),
 				messages: convertToR1Format([{ role: "user", content: systemPrompt }, ...messages]),
 			})
 
@@ -85,7 +98,7 @@ export class FeatherlessHandler extends BaseOpenAiCompatibleProvider<Featherless
 				yield processedChunk
 			}
 		} else {
-			yield* super.createMessage(systemPrompt, messages)
+			yield* super.createMessage(systemPrompt, messages, metadata)
 		}
 	}
 

src/api/providers/gemini.ts

@@ -209,7 +209,7 @@ export class GeminiHandler extends BaseProvider implements SingleCompletionHandl
 		return citationLinks.join(", ")
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		try {
 			const { id: model } = this.getModel()
 

src/api/providers/glama.ts

@@ -116,13 +116,15 @@ export class GlamaHandler extends RouterProvider implements SingleCompletionHand
 		}
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		const { id: modelId, info } = await this.fetchModel()
 
 		try {
 			const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsNonStreaming = {
 				model: modelId,
 				messages: [{ role: "user", content: prompt }],
+				prompt_cache_key: metadata?.taskId,
+				safety_identifier: metadata?.safetyIdentifier,
 			}
 
 			if (this.supportsTemperature(modelId)) {

src/api/providers/huggingface.ts

@@ -59,6 +59,8 @@ export class HuggingFaceHandler extends BaseProvider implements SingleCompletion
 			messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
 			stream: true,
 			stream_options: { include_usage: true },
+			prompt_cache_key: metadata?.taskId,
+			safety_identifier: metadata?.safetyIdentifier,
 		}
 
 		// Add max_tokens if specified
@@ -93,13 +95,15 @@ export class HuggingFaceHandler extends BaseProvider implements SingleCompletion
 		}
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		const modelId = this.options.huggingFaceModelId || "meta-llama/Llama-3.3-70B-Instruct"
 
 		try {
 			const response = await this.client.chat.completions.create({
 				model: modelId,
 				messages: [{ role: "user", content: prompt }],
+				prompt_cache_key: metadata?.taskId,
+				safety_identifier: metadata?.safetyIdentifier,
 			})
 
 			return response.choices[0]?.message.content || ""

src/api/providers/human-relay.ts

@@ -82,7 +82,7 @@ export class HumanRelayHandler implements ApiHandler, SingleCompletionHandler {
 	 * Implementation of a single prompt
 	 * @param prompt Prompt content
 	 */
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		// Copy to clipboard
 		await vscode.env.clipboard.writeText(prompt)
 

src/api/providers/lite-llm.ts

@@ -123,6 +123,8 @@ export class LiteLLMHandler extends RouterProvider implements SingleCompletionHa
 			stream_options: {
 				include_usage: true,
 			},
+			prompt_cache_key: metadata?.taskId,
+			safety_identifier: metadata?.safetyIdentifier,
 		}
 
 		// GPT-5 models require max_completion_tokens instead of the deprecated max_tokens parameter
@@ -191,7 +193,7 @@ export class LiteLLMHandler extends RouterProvider implements SingleCompletionHa
 		}
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		const { id: modelId, info } = await this.fetchModel()
 
 		// Check if this is a GPT-5 model that requires max_completion_tokens instead of max_tokens
@@ -201,6 +203,8 @@ export class LiteLLMHandler extends RouterProvider implements SingleCompletionHa
 			const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsNonStreaming = {
 				model: modelId,
 				messages: [{ role: "user", content: prompt }],
+				prompt_cache_key: metadata?.taskId,
+				safety_identifier: metadata?.safetyIdentifier,
 			}
 
 			if (this.supportsTemperature(modelId)) {

src/api/providers/lm-studio.ts

@@ -87,6 +87,8 @@ export class LmStudioHandler extends BaseProvider implements SingleCompletionHan
 				messages: openAiMessages,
 				temperature: this.options.modelTemperature ?? LMSTUDIO_DEFAULT_TEMPERATURE,
 				stream: true,
+				prompt_cache_key: metadata?.taskId,
+				safety_identifier: metadata?.safetyIdentifier,
 			}
 
 			if (this.options.lmStudioSpeculativeDecodingEnabled && this.options.lmStudioDraftModelId) {
@@ -159,14 +161,16 @@ export class LmStudioHandler extends BaseProvider implements SingleCompletionHan
 		}
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		try {
 			// Create params object with optional draft model
 			const params: any = {
 				model: this.getModel().id,
 				messages: [{ role: "user", content: prompt }],
 				temperature: this.options.modelTemperature ?? LMSTUDIO_DEFAULT_TEMPERATURE,
 				stream: false,
+				prompt_cache_key: metadata?.taskId,
+				safety_identifier: metadata?.safetyIdentifier,
 			}
 
 			// Add draft model if speculative decoding is enabled and a draft model is specified

src/api/providers/mistral.ts

@@ -104,7 +104,7 @@ export class MistralHandler extends BaseProvider implements SingleCompletionHand
 		return { id, info, maxTokens, temperature }
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		try {
 			const { id: model, temperature } = this.getModel()
 

src/api/providers/native-ollama.ts

@@ -280,7 +280,7 @@ export class NativeOllamaHandler extends BaseProvider implements SingleCompletio
 		}
 	}
 
-	async completePrompt(prompt: string): Promise<string> {
+	async completePrompt(prompt: string, metadata?: ApiHandlerCreateMessageMetadata): Promise<string> {
 		try {
 			const client = this.ensureClient()
 			const { id: modelId } = await this.fetchModel()