feat: add native tool support for LiteLLM provider (#9719)

Author: daniel-lxs

packages/types/src/providers/lite-llm.ts

@@ -8,6 +8,7 @@ export const litellmDefaultModelInfo: ModelInfo = {
 	contextWindow: 200_000,
 	supportsImages: true,
 	supportsPromptCache: true,
+	supportsNativeTools: true,
 	inputPrice: 3.0,
 	outputPrice: 15.0,
 	cacheWritesPrice: 3.75,

src/api/providers/fetchers/__tests__/litellm.spec.ts

@@ -222,17 +222,23 @@ describe("getLiteLLMModels", () => {
 				contextWindow: 200000,
 				supportsImages: true,
 				supportsPromptCache: false,
+				supportsNativeTools: false,
 				inputPrice: 3,
 				outputPrice: 15,
+				cacheWritesPrice: undefined,
+				cacheReadsPrice: undefined,
 				description: "claude-3-5-sonnet via LiteLLM proxy",
 			},
 			"gpt-4-turbo": {
 				maxTokens: 8192,
 				contextWindow: 128000,
 				supportsImages: false,
 				supportsPromptCache: false,
+				supportsNativeTools: false,
 				inputPrice: 10,
 				outputPrice: 30,
+				cacheWritesPrice: undefined,
+				cacheReadsPrice: undefined,
 				description: "gpt-4-turbo via LiteLLM proxy",
 			},
 		})
@@ -299,8 +305,11 @@ describe("getLiteLLMModels", () => {
 			contextWindow: 200000,
 			supportsImages: true,
 			supportsPromptCache: false,
+			supportsNativeTools: false,
 			inputPrice: undefined,
 			outputPrice: undefined,
+			cacheWritesPrice: undefined,
+			cacheReadsPrice: undefined,
 			description: "test-computer-model via LiteLLM proxy",
 		})
 
@@ -309,8 +318,11 @@ describe("getLiteLLMModels", () => {
 			contextWindow: 200000,
 			supportsImages: false,
 			supportsPromptCache: false,
+			supportsNativeTools: false,
 			inputPrice: undefined,
 			outputPrice: undefined,
+			cacheWritesPrice: undefined,
+			cacheReadsPrice: undefined,
 			description: "test-non-computer-model via LiteLLM proxy",
 		})
 	})
@@ -443,8 +455,11 @@ describe("getLiteLLMModels", () => {
 			contextWindow: 200000,
 			supportsImages: true,
 			supportsPromptCache: false,
+			supportsNativeTools: false,
 			inputPrice: undefined,
 			outputPrice: undefined,
+			cacheWritesPrice: undefined,
+			cacheReadsPrice: undefined,
 			description: "claude-3-5-sonnet-latest via LiteLLM proxy",
 		})
 
@@ -453,8 +468,11 @@ describe("getLiteLLMModels", () => {
 			contextWindow: 128000,
 			supportsImages: false,
 			supportsPromptCache: false,
+			supportsNativeTools: false,
 			inputPrice: undefined,
 			outputPrice: undefined,
+			cacheWritesPrice: undefined,
+			cacheReadsPrice: undefined,
 			description: "gpt-4-turbo via LiteLLM proxy",
 		})
 	})
@@ -515,8 +533,11 @@ describe("getLiteLLMModels", () => {
 			contextWindow: 200000,
 			supportsImages: true,
 			supportsPromptCache: false,
+			supportsNativeTools: false,
 			inputPrice: undefined,
 			outputPrice: undefined,
+			cacheWritesPrice: undefined,
+			cacheReadsPrice: undefined,
 			description: "claude-3-5-sonnet-latest via LiteLLM proxy",
 		})
 
@@ -525,8 +546,11 @@ describe("getLiteLLMModels", () => {
 			contextWindow: 128000,
 			supportsImages: false,
 			supportsPromptCache: false,
+			supportsNativeTools: false,
 			inputPrice: undefined,
 			outputPrice: undefined,
+			cacheWritesPrice: undefined,
+			cacheReadsPrice: undefined,
 			description: "custom-model via LiteLLM proxy",
 		})
 
@@ -535,8 +559,11 @@ describe("getLiteLLMModels", () => {
 			contextWindow: 128000,
 			supportsImages: false,
 			supportsPromptCache: false,
+			supportsNativeTools: false,
 			inputPrice: undefined,
 			outputPrice: undefined,
+			cacheWritesPrice: undefined,
+			cacheReadsPrice: undefined,
 			description: "another-custom-model via LiteLLM proxy",
 		})
 	})
@@ -646,6 +673,7 @@ describe("getLiteLLMModels", () => {
 			contextWindow: 200000,
 			supportsImages: true,
 			supportsPromptCache: false,
+			supportsNativeTools: false,
 			inputPrice: undefined,
 			outputPrice: undefined,
 			cacheWritesPrice: undefined,
@@ -659,6 +687,7 @@ describe("getLiteLLMModels", () => {
 			contextWindow: 128000,
 			supportsImages: false,
 			supportsPromptCache: false,
+			supportsNativeTools: false,
 			inputPrice: undefined,
 			outputPrice: undefined,
 			cacheWritesPrice: undefined,
@@ -672,6 +701,7 @@ describe("getLiteLLMModels", () => {
 			contextWindow: 100000,
 			supportsImages: false,
 			supportsPromptCache: false,
+			supportsNativeTools: false,
 			inputPrice: undefined,
 			outputPrice: undefined,
 			cacheWritesPrice: undefined,

src/api/providers/fetchers/litellm.ts

@@ -45,6 +45,11 @@ export async function getLiteLLMModels(apiKey: string, baseUrl: string): Promise
 					contextWindow: modelInfo.max_input_tokens || 200000,
 					supportsImages: Boolean(modelInfo.supports_vision),
 					supportsPromptCache: Boolean(modelInfo.supports_prompt_caching),
+					supportsNativeTools: Boolean(
+						modelInfo.supports_function_calling ||
+							modelInfo.supports_tool_choice ||
+							modelInfo.supports_tool_use,
+					),
 					inputPrice: modelInfo.input_cost_per_token ? modelInfo.input_cost_per_token * 1000000 : undefined,
 					outputPrice: modelInfo.output_cost_per_token
 						? modelInfo.output_cost_per_token * 1000000

src/api/providers/lite-llm.ts

@@ -1,7 +1,7 @@
 import OpenAI from "openai"
 import { Anthropic } from "@anthropic-ai/sdk" // Keep for type usage only
 
-import { litellmDefaultModelId, litellmDefaultModelInfo } from "@roo-code/types"
+import { litellmDefaultModelId, litellmDefaultModelInfo, TOOL_PROTOCOL } from "@roo-code/types"
 
 import { calculateApiCostOpenAI } from "../../shared/cost"
 
@@ -116,13 +116,24 @@ export class LiteLLMHandler extends RouterProvider implements SingleCompletionHa
 		// Check if this is a GPT-5 model that requires max_completion_tokens instead of max_tokens
 		const isGPT5Model = this.isGpt5(modelId)
 
+		// Check if model supports native tools and tools are provided with native protocol
+		const supportsNativeTools = info.supportsNativeTools ?? false
+		const useNativeTools =
+			supportsNativeTools &&
+			metadata?.tools &&
+			metadata.tools.length > 0 &&
+			metadata?.toolProtocol === TOOL_PROTOCOL.NATIVE
+
 		const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {
 			model: modelId,
 			messages: [systemMessage, ...enhancedMessages],
 			stream: true,
 			stream_options: {
 				include_usage: true,
 			},
+			...(useNativeTools && { tools: this.convertToolsForOpenAI(metadata.tools) }),
+			...(useNativeTools && metadata.tool_choice && { tool_choice: metadata.tool_choice }),
+			...(useNativeTools && { parallel_tool_calls: metadata?.parallelToolCalls ?? false }),
 		}
 
 		// GPT-5 models require max_completion_tokens instead of the deprecated max_tokens parameter
@@ -149,6 +160,19 @@ export class LiteLLMHandler extends RouterProvider implements SingleCompletionHa
 					yield { type: "text", text: delta.content }
 				}
 
+				// Handle tool calls in stream - emit partial chunks for NativeToolCallParser
+				if (delta?.tool_calls) {
+					for (const toolCall of delta.tool_calls) {
+						yield {
+							type: "tool_call_partial",
+							index: toolCall.index,
+							id: toolCall.id,
+							name: toolCall.function?.name,
+							arguments: toolCall.function?.arguments,
+						}
+					}
+				}
+
 				if (usage) {
 					lastUsage = usage
 				}