[Fix] NanoGPT Muse Spark fails during tool use (#1310)

* fix(nanogpt): serialize Muse Spark tool calls

* fix(nanogpt): preserve Muse tool-result context

---------

Co-authored-by: @navedmerchant <14171946+navedmerchant@users.noreply.github.com>

Author: zoomote[bot]

src/api/providers/__tests__/nanogpt.spec.ts

@@ -171,6 +171,78 @@ describe("NanoGptHandler", () => {
 		expect(mockCreate.mock.calls[0][0]).not.toHaveProperty("max_completion_tokens")
 	})
 
+	it("keeps Muse Spark tool-result history contiguous across turns", async () => {
+		const modelId = "meta/muse-spark-1.2-contributor"
+		vi.mocked(getModels).mockResolvedValue({
+			[modelId]: {
+				maxTokens: 65_536,
+				contextWindow: 1_000_000,
+				supportsPromptCache: false,
+			},
+		})
+		const tools: OpenAI.Chat.ChatCompletionTool[] = [
+			{ type: "function", function: { name: "read_file", parameters: { type: "object" } } },
+		]
+		const toolHistory: Anthropic.Messages.MessageParam[] = [
+			{
+				role: "assistant",
+				content: [{ type: "tool_use", id: "call_1", name: "read_file", input: { path: "first.txt" } }],
+			},
+			{
+				role: "user",
+				content: [
+					{ type: "tool_result", tool_use_id: "call_1", content: "first result" },
+					{ type: "text", text: "<environment_details>first context</environment_details>" },
+				],
+			},
+			{
+				role: "assistant",
+				content: [{ type: "tool_use", id: "call_2", name: "read_file", input: { path: "second.txt" } }],
+			},
+			{
+				role: "user",
+				content: [
+					{ type: "tool_result", tool_use_id: "call_2", content: "second result" },
+					{ type: "text", text: "<environment_details>second context</environment_details>" },
+				],
+			},
+		]
+
+		await collectStream(
+			new NanoGptHandler({ nanoGptModelId: modelId }).createMessage("sys", toolHistory, {
+				taskId: "task",
+				tools,
+				tool_choice: "auto",
+				parallelToolCalls: true,
+			}),
+		)
+
+		expect(mockCreate).toHaveBeenCalledWith(
+			expect.objectContaining({
+				model: modelId,
+				messages: [
+					{ role: "system", content: "sys" },
+					expect.objectContaining({ role: "assistant", tool_calls: [expect.anything()] }),
+					{
+						role: "tool",
+						tool_call_id: "call_1",
+						content: "first result\n\n<environment_details>first context</environment_details>",
+					},
+					expect.objectContaining({ role: "assistant", tool_calls: [expect.anything()] }),
+					{
+						role: "tool",
+						tool_call_id: "call_2",
+						content: "second result\n\n<environment_details>second context</environment_details>",
+					},
+				],
+				tools: [expect.objectContaining({ function: expect.objectContaining({ name: "read_file" }) })],
+				tool_choice: "auto",
+				parallel_tool_calls: true,
+			}),
+			expect.anything(),
+		)
+	})
+
 	it("omits temperature when it was not explicitly configured", async () => {
 		await collectStream(new NanoGptHandler({ nanoGptModelId: "model:thinking" }).createMessage("sys", messages))
 		expect(mockCreate.mock.calls[0][0]).not.toHaveProperty("temperature")

src/api/providers/nanogpt.ts

@@ -27,6 +27,8 @@ type NanoGptUsage = OpenAI.CompletionUsage & {
 
 type NanoGptCachingRequest = { caching?: true }
 
+const NANO_GPT_MERGED_TOOL_RESULT_MODELS = new Set(["meta/muse-spark-1.2-contributor"])
+
 const OPENAI_REASONING_EFFORTS = ["low", "medium", "high"] as const
 type OpenAiReasoningEffort = (typeof OPENAI_REASONING_EFFORTS)[number]
 
@@ -93,7 +95,12 @@ export class NanoGptHandler extends RouterProvider implements SingleCompletionHa
 		const { id: canonicalModelId, info } = await this.fetchModel()
 		const body: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming & NanoGptCachingRequest = {
 			model: this.getRequestModelId(canonicalModelId),
-			messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
+			messages: [
+				{ role: "system", content: systemPrompt },
+				...convertToOpenAiMessages(messages, {
+					mergeToolResultText: NANO_GPT_MERGED_TOOL_RESULT_MODELS.has(canonicalModelId),
+				}),
+			],
 			stream: true,
 			stream_options: { include_usage: true },
 			max_tokens: info.maxTokens ?? undefined,