chore: import upstream snapshot with attribution

apps/sim/providers/fireworks/index.test.ts

@@ -0,0 +1,232 @@
+/**
+ * @vitest-environment node
+ */
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+
+const {
+  mockCreate,
+  mockSupportsNativeStructuredOutputs,
+  mockPrepareToolsWithUsageControl,
+  mockExecuteTool,
+} = vi.hoisted(() => ({
+  mockCreate: vi.fn(),
+  mockSupportsNativeStructuredOutputs: vi.fn(),
+  mockPrepareToolsWithUsageControl: vi.fn(),
+  mockExecuteTool: vi.fn(),
+}))
+
+vi.mock('openai', () => ({
+  default: vi.fn().mockImplementation(
+    class {
+      chat = { completions: { create: mockCreate } }
+    }
+  ),
+}))
+
+vi.mock('@/providers', () => ({ MAX_TOOL_ITERATIONS: 5 }))
+
+vi.mock('@/providers/models', () => ({
+  getProviderFileAttachment: vi
+    .fn()
+    .mockReturnValue({ maxBytes: 10 * 1024 * 1024, strategy: 'inline' }),
+  INLINE_ATTACHMENT_MAX_BYTES: 10 * 1024 * 1024,
+  getProviderModels: vi.fn().mockReturnValue([]),
+  getProviderDefaultModel: vi.fn().mockReturnValue('llama-v3p1-70b-instruct'),
+}))
+
+vi.mock('@/providers/attachments', () => ({
+  formatMessagesForProvider: vi.fn((messages) => messages),
+}))
+
+vi.mock('@/providers/fireworks/utils', () => ({
+  supportsNativeStructuredOutputs: mockSupportsNativeStructuredOutputs,
+  createReadableStreamFromOpenAIStream: vi.fn(() => ({}) as ReadableStream),
+  checkForForcedToolUsage: vi.fn(() => ({ hasUsedForcedTool: false, usedForcedTools: [] })),
+}))
+
+vi.mock('@/providers/trace-enrichment', () => ({
+  enrichLastModelSegmentFromChatCompletions: vi.fn(),
+}))
+
+vi.mock('@/providers/utils', () => ({
+  calculateCost: vi.fn().mockReturnValue({ input: 0, output: 0, total: 0 }),
+  generateSchemaInstructions: vi.fn(() => 'SCHEMA_INSTRUCTIONS'),
+  prepareToolExecution: vi.fn(() => ({ toolParams: { x: 1 }, executionParams: { x: 1 } })),
+  prepareToolsWithUsageControl: mockPrepareToolsWithUsageControl,
+  sumToolCosts: vi.fn().mockReturnValue(0),
+}))
+
+vi.mock('@/tools', () => ({ executeTool: mockExecuteTool }))
+
+import { fireworksProvider } from '@/providers/fireworks/index'
+import { ProviderError } from '@/providers/types'
+
+const textResponse = (content: string) => ({
+  choices: [{ message: { content, tool_calls: [] } }],
+  usage: { prompt_tokens: 10, completion_tokens: 5, total_tokens: 15 },
+})
+
+const toolCallResponse = () => ({
+  choices: [
+    {
+      message: {
+        content: null,
+        tool_calls: [
+          { id: 'call_1', type: 'function', function: { name: 'my_tool', arguments: '{"x":1}' } },
+        ],
+      },
+    },
+  ],
+  usage: { prompt_tokens: 8, completion_tokens: 4, total_tokens: 12 },
+})
+
+const toolDef = {
+  id: 'my_tool',
+  name: 'my_tool',
+  description: '',
+  params: {},
+  parameters: { type: 'object', properties: {}, required: [] },
+}
+
+const callBody = (index: number) => mockCreate.mock.calls[index][0]
+const lastCallBody = () => mockCreate.mock.calls.at(-1)?.[0]
+
+describe('fireworksProvider', () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+    mockSupportsNativeStructuredOutputs.mockResolvedValue(true)
+    mockPrepareToolsWithUsageControl.mockImplementation((tools) => ({
+      tools,
+      toolChoice: 'auto',
+      forcedTools: [],
+    }))
+    mockExecuteTool.mockResolvedValue({ success: true, output: { ok: true } })
+  })
+
+  const baseRequest = {
+    model: 'fireworks/llama-v3p1-70b-instruct',
+    systemPrompt: 'You are helpful.',
+    messages: [{ role: 'user' as const, content: 'Hello' }],
+    apiKey: 'fw-test-key',
+  }
+
+  it('throws when the API key is missing', async () => {
+    await expect(
+      fireworksProvider.executeRequest({ ...baseRequest, apiKey: undefined })
+    ).rejects.toThrow('API key is required for Fireworks')
+  })
+
+  it('returns content and token usage for a simple request', async () => {
+    mockCreate.mockResolvedValueOnce(textResponse('hi there'))
+
+    const result = await fireworksProvider.executeRequest(baseRequest)
+
+    expect(result).toMatchObject({
+      content: 'hi there',
+      model: 'llama-v3p1-70b-instruct',
+      tokens: { input: 10, output: 5, total: 15 },
+    })
+  })
+
+  it('wraps API errors in a ProviderError', async () => {
+    mockCreate.mockRejectedValueOnce(new Error('boom'))
+
+    await expect(fireworksProvider.executeRequest(baseRequest)).rejects.toBeInstanceOf(
+      ProviderError
+    )
+  })
+
+  it('streams directly when there are no tools', async () => {
+    mockCreate.mockResolvedValueOnce({})
+
+    const result = await fireworksProvider.executeRequest({ ...baseRequest, stream: true })
+
+    expect(lastCallBody()).toMatchObject({ stream: true, stream_options: { include_usage: true } })
+    expect(result).toHaveProperty('stream')
+    expect(result).toHaveProperty('execution')
+  })
+
+  it('sends a json_schema response_format with no strict field', async () => {
+    mockCreate.mockResolvedValueOnce(textResponse('{}'))
+
+    await fireworksProvider.executeRequest({
+      ...baseRequest,
+      responseFormat: { name: 'my_schema', schema: { type: 'object' }, strict: true },
+    })
+
+    expect(lastCallBody().response_format).toEqual({
+      type: 'json_schema',
+      json_schema: { name: 'my_schema', schema: { type: 'object' } },
+    })
+    expect(lastCallBody().response_format.json_schema).not.toHaveProperty('strict')
+  })
+
+  it('falls back to json_object with prompt instructions when native is unsupported', async () => {
+    mockSupportsNativeStructuredOutputs.mockResolvedValue(false)
+    mockCreate.mockResolvedValueOnce(textResponse('{}'))
+
+    await fireworksProvider.executeRequest({
+      ...baseRequest,
+      responseFormat: { name: 'my_schema', schema: { type: 'object' } },
+    })
+
+    expect(lastCallBody().response_format).toEqual({ type: 'json_object' })
+    expect(lastCallBody().messages.at(-1)).toEqual({
+      role: 'user',
+      content: 'SCHEMA_INSTRUCTIONS',
+    })
+  })
+
+  it('defers response_format to a final call when tools are active', async () => {
+    mockCreate
+      .mockResolvedValueOnce(textResponse('intermediate'))
+      .mockResolvedValueOnce(textResponse('{"done":true}'))
+
+    await fireworksProvider.executeRequest({
+      ...baseRequest,
+      responseFormat: { name: 'my_schema', schema: { type: 'object' } },
+      tools: [toolDef],
+    })
+
+    expect(mockCreate).toHaveBeenCalledTimes(2)
+    expect(callBody(0).response_format).toBeUndefined()
+    expect(callBody(0).tools).toBeDefined()
+    expect(callBody(1).response_format).toEqual({
+      type: 'json_schema',
+      json_schema: { name: 'my_schema', schema: { type: 'object' } },
+    })
+    expect(callBody(1).tools).toBeUndefined()
+  })
+
+  it('runs the tool loop and threads tool results back into the conversation', async () => {
+    mockCreate
+      .mockResolvedValueOnce(toolCallResponse())
+      .mockResolvedValueOnce(textResponse('final answer'))
+
+    const result = await fireworksProvider.executeRequest({ ...baseRequest, tools: [toolDef] })
+
+    expect(mockExecuteTool).toHaveBeenCalledWith('my_tool', { x: 1 }, expect.anything())
+    expect(result).toMatchObject({ content: 'final answer' })
+    expect((result as { toolCalls?: unknown[] }).toolCalls).toHaveLength(1)
+
+    const followUpMessages = callBody(1).messages
+    expect(followUpMessages).toContainEqual(
+      expect.objectContaining({ role: 'assistant', tool_calls: expect.any(Array) })
+    )
+    expect(followUpMessages).toContainEqual(
+      expect.objectContaining({ role: 'tool', tool_call_id: 'call_1' })
+    )
+  })
+
+  it("forces tool_choice 'none' on the final streaming call after tools run", async () => {
+    mockCreate
+      .mockResolvedValueOnce(toolCallResponse())
+      .mockResolvedValueOnce(textResponse('done'))
+      .mockResolvedValueOnce({})
+
+    await fireworksProvider.executeRequest({ ...baseRequest, stream: true, tools: [toolDef] })
+
+    expect(mockCreate).toHaveBeenCalledTimes(3)
+    expect(lastCallBody()).toMatchObject({ tool_choice: 'none', stream: true })
+  })
+})

apps/sim/providers/fireworks/index.ts

@@ -0,0 +1,599 @@
+import { createLogger } from '@sim/logger'
+import { getErrorMessage, toError } from '@sim/utils/errors'
+import OpenAI from 'openai'
+import type { ChatCompletionCreateParamsStreaming } from 'openai/resources/chat/completions'
+import type { StreamingExecution } from '@/executor/types'
+import { MAX_TOOL_ITERATIONS } from '@/providers'
+import { formatMessagesForProvider } from '@/providers/attachments'
+import {
+  checkForForcedToolUsage,
+  createReadableStreamFromOpenAIStream,
+  supportsNativeStructuredOutputs,
+} from '@/providers/fireworks/utils'
+import { getProviderDefaultModel, getProviderModels } from '@/providers/models'
+import { createStreamingExecution } from '@/providers/streaming-execution'
+import { adaptOpenAIChatToolSchema } from '@/providers/tool-schema-adapter'
+import { enrichLastModelSegmentFromChatCompletions } from '@/providers/trace-enrichment'
+import type {
+  FunctionCallResponse,
+  Message,
+  ProviderConfig,
+  ProviderRequest,
+  ProviderResponse,
+  TimeSegment,
+} from '@/providers/types'
+import { ProviderError } from '@/providers/types'
+import {
+  calculateCost,
+  generateSchemaInstructions,
+  prepareToolExecution,
+  prepareToolsWithUsageControl,
+  sumToolCosts,
+} from '@/providers/utils'
+import { executeTool } from '@/tools'
+
+const logger = createLogger('FireworksProvider')
+
+/**
+ * Applies structured output configuration to a payload based on model capabilities.
+ * Uses native json_schema for supported models, falls back to json_object with prompt instructions.
+ */
+async function applyResponseFormat(
+  targetPayload: any,
+  messages: any[],
+  responseFormat: any,
+  model: string
+): Promise<any[]> {
+  const useNative = await supportsNativeStructuredOutputs(model)
+
+  if (useNative) {
+    logger.info('Using native structured outputs for Fireworks model', { model })
+    targetPayload.response_format = {
+      type: 'json_schema',
+      json_schema: {
+        name: responseFormat.name || 'response_schema',
+        schema: responseFormat.schema || responseFormat,
+      },
+    }
+    return messages
+  }
+
+  logger.info('Using json_object mode with prompt instructions for Fireworks model', { model })
+  const schema = responseFormat.schema || responseFormat
+  const schemaInstructions = generateSchemaInstructions(schema, responseFormat.name)
+  targetPayload.response_format = { type: 'json_object' }
+  return [...messages, { role: 'user', content: schemaInstructions }]
+}
+
+export const fireworksProvider: ProviderConfig = {
+  id: 'fireworks',
+  name: 'Fireworks',
+  description: 'Fast inference for open-source models via Fireworks AI',
+  version: '1.0.0',
+  models: getProviderModels('fireworks'),
+  defaultModel: getProviderDefaultModel('fireworks'),
+
+  executeRequest: async (
+    request: ProviderRequest
+  ): Promise<ProviderResponse | StreamingExecution> => {
+    if (!request.apiKey) {
+      throw new Error('API key is required for Fireworks')
+    }
+
+    const client = new OpenAI({
+      apiKey: request.apiKey,
+      baseURL: 'https://api.fireworks.ai/inference/v1',
+    })
+
+    const requestedModel = request.model.replace(/^fireworks\//, '')
+
+    logger.info('Preparing Fireworks request', {
+      model: requestedModel,
+      hasSystemPrompt: !!request.systemPrompt,
+      hasMessages: !!request.messages?.length,
+      hasTools: !!request.tools?.length,
+      toolCount: request.tools?.length || 0,
+      hasResponseFormat: !!request.responseFormat,
+      stream: !!request.stream,
+    })
+
+    const allMessages: Message[] = []
+
+    if (request.systemPrompt) {
+      allMessages.push({ role: 'system', content: request.systemPrompt })
+    }
+
+    if (request.context) {
+      allMessages.push({ role: 'user', content: request.context })
+    }
+
+    if (request.messages) {
+      allMessages.push(...request.messages)
+    }
+    const formattedMessages = formatMessagesForProvider(allMessages, 'fireworks') as Message[]
+
+    const tools = request.tools?.length
+      ? request.tools.map((tool) => adaptOpenAIChatToolSchema(tool))
+      : undefined
+
+    const payload: any = {
+      model: requestedModel,
+      messages: formattedMessages,
+    }
+
+    if (request.temperature !== undefined) payload.temperature = request.temperature
+    if (request.maxTokens != null) payload.max_tokens = request.maxTokens
+
+    let preparedTools: ReturnType<typeof prepareToolsWithUsageControl> | null = null
+    let hasActiveTools = false
+    if (tools?.length) {
+      preparedTools = prepareToolsWithUsageControl(tools, request.tools, logger, 'fireworks')
+      const { tools: filteredTools, toolChoice } = preparedTools
+      if (filteredTools?.length && toolChoice) {
+        payload.tools = filteredTools
+        payload.tool_choice = toolChoice
+        hasActiveTools = true
+      }
+    }
+
+    const providerStartTime = Date.now()
+    const providerStartTimeISO = new Date(providerStartTime).toISOString()
+
+    try {
+      if (request.responseFormat && !hasActiveTools) {
+        payload.messages = await applyResponseFormat(
+          payload,
+          payload.messages,
+          request.responseFormat,
+          requestedModel
+        )
+      }
+
+      if (request.stream && (!tools || tools.length === 0 || !hasActiveTools)) {
+        const streamingParams: ChatCompletionCreateParamsStreaming = {
+          ...payload,
+          stream: true,
+          stream_options: { include_usage: true },
+        }
+        const streamResponse = await client.chat.completions.create(
+          streamingParams,
+          request.abortSignal ? { signal: request.abortSignal } : undefined
+        )
+
+        const streamingResult = createStreamingExecution({
+          model: requestedModel,
+          providerStartTime,
+          providerStartTimeISO,
+          timing: { kind: 'simple', segmentName: request.model },
+          initialTokens: { input: 0, output: 0, total: 0 },
+          initialCost: { input: 0, output: 0, total: 0 },
+          createStream: ({ output, finalizeTiming }) =>
+            createReadableStreamFromOpenAIStream(streamResponse, (content, usage) => {
+              output.content = content
+              output.tokens = {
+                input: usage.prompt_tokens,
+                output: usage.completion_tokens,
+                total: usage.total_tokens,
+              }
+
+              const costResult = calculateCost(
+                requestedModel,
+                usage.prompt_tokens,
+                usage.completion_tokens
+              )
+              output.cost = {
+                input: costResult.input,
+                output: costResult.output,
+                total: costResult.total,
+              }
+
+              finalizeTiming()
+            }),
+        })
+
+        return streamingResult
+      }
+
+      const initialCallTime = Date.now()
+      const originalToolChoice = payload.tool_choice
+      const forcedTools = preparedTools?.forcedTools || []
+      let usedForcedTools: string[] = []
+
+      let currentResponse = await client.chat.completions.create(
+        payload,
+        request.abortSignal ? { signal: request.abortSignal } : undefined
+      )
+      const firstResponseTime = Date.now() - initialCallTime
+
+      let content = currentResponse.choices[0]?.message?.content || ''
+      const tokens = {
+        input: currentResponse.usage?.prompt_tokens || 0,
+        output: currentResponse.usage?.completion_tokens || 0,
+        total: currentResponse.usage?.total_tokens || 0,
+      }
+      const toolCalls: FunctionCallResponse[] = []
+      const toolResults: Record<string, unknown>[] = []
+      const currentMessages = [...formattedMessages]
+      let iterationCount = 0
+      let modelTime = firstResponseTime
+      let toolsTime = 0
+      let hasUsedForcedTool = false
+      const timeSegments: TimeSegment[] = [
+        {
+          type: 'model',
+          name: request.model,
+          startTime: initialCallTime,
+          endTime: initialCallTime + firstResponseTime,
+          duration: firstResponseTime,
+        },
+      ]
+
+      const forcedToolResult = checkForForcedToolUsage(
+        currentResponse,
+        originalToolChoice,
+        forcedTools,
+        usedForcedTools
+      )
+      hasUsedForcedTool = forcedToolResult.hasUsedForcedTool
+      usedForcedTools = forcedToolResult.usedForcedTools
+
+      while (iterationCount < MAX_TOOL_ITERATIONS) {
+        if (currentResponse.choices[0]?.message?.content) {
+          content = currentResponse.choices[0].message.content
+        }
+
+        const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
+
+        enrichLastModelSegmentFromChatCompletions(
+          timeSegments,
+          currentResponse,
+          toolCallsInResponse,
+          { model: request.model, provider: 'fireworks' }
+        )
+
+        if (!toolCallsInResponse || toolCallsInResponse.length === 0) {
+          break
+        }
+
+        const toolsStartTime = Date.now()
+
+        const toolExecutionPromises = toolCallsInResponse.map(async (toolCall) => {
+          const toolCallStartTime = Date.now()
+          const toolName = toolCall.function.name
+
+          try {
+            const toolArgs = JSON.parse(toolCall.function.arguments)
+            const tool = request.tools?.find((t) => t.id === toolName)
+
+            if (!tool) return null
+
+            const { toolParams, executionParams } = prepareToolExecution(tool, toolArgs, request)
+            const result = await executeTool(toolName, executionParams, {
+              signal: request.abortSignal,
+            })
+            const toolCallEndTime = Date.now()
+
+            return {
+              toolCall,
+              toolName,
+              toolParams,
+              result,
+              startTime: toolCallStartTime,
+              endTime: toolCallEndTime,
+              duration: toolCallEndTime - toolCallStartTime,
+            }
+          } catch (error) {
+            const toolCallEndTime = Date.now()
+            logger.error('Error processing tool call (Fireworks):', {
+              error: toError(error).message,
+              toolName,
+            })
+
+            return {
+              toolCall,
+              toolName,
+              toolParams: {},
+              result: {
+                success: false,
+                output: undefined,
+                error: getErrorMessage(error, 'Tool execution failed'),
+              },
+              startTime: toolCallStartTime,
+              endTime: toolCallEndTime,
+              duration: toolCallEndTime - toolCallStartTime,
+            }
+          }
+        })
+
+        const executionResults = await Promise.allSettled(toolExecutionPromises)
+
+        currentMessages.push({
+          role: 'assistant',
+          content: null,
+          tool_calls: toolCallsInResponse.map((tc) => ({
+            id: tc.id,
+            type: 'function',
+            function: {
+              name: tc.function.name,
+              arguments: tc.function.arguments,
+            },
+          })),
+        })
+
+        for (const settledResult of executionResults) {
+          if (settledResult.status === 'rejected' || !settledResult.value) continue
+
+          const { toolCall, toolName, toolParams, result, startTime, endTime, duration } =
+            settledResult.value
+
+          timeSegments.push({
+            type: 'tool',
+            name: toolName,
+            startTime: startTime,
+            endTime: endTime,
+            duration: duration,
+            toolCallId: toolCall.id,
+          })
+
+          let resultContent: any
+          if (result.success) {
+            toolResults.push(result.output!)
+            resultContent = result.output
+          } else {
+            resultContent = {
+              error: true,
+              message: result.error || 'Tool execution failed',
+              tool: toolName,
+            }
+          }
+
+          toolCalls.push({
+            name: toolName,
+            arguments: toolParams,
+            startTime: new Date(startTime).toISOString(),
+            endTime: new Date(endTime).toISOString(),
+            duration: duration,
+            result: resultContent,
+            success: result.success,
+          })
+
+          currentMessages.push({
+            role: 'tool',
+            tool_call_id: toolCall.id,
+            content: JSON.stringify(resultContent),
+          })
+        }
+
+        const thisToolsTime = Date.now() - toolsStartTime
+        toolsTime += thisToolsTime
+
+        const nextPayload = {
+          ...payload,
+          messages: currentMessages,
+        }
+
+        if (typeof originalToolChoice === 'object' && hasUsedForcedTool && forcedTools.length > 0) {
+          const remainingTools = forcedTools.filter((tool) => !usedForcedTools.includes(tool))
+          if (remainingTools.length > 0) {
+            nextPayload.tool_choice = { type: 'function', function: { name: remainingTools[0] } }
+          } else {
+            nextPayload.tool_choice = 'auto'
+          }
+        }
+
+        const nextModelStartTime = Date.now()
+        currentResponse = await client.chat.completions.create(
+          nextPayload,
+          request.abortSignal ? { signal: request.abortSignal } : undefined
+        )
+        const nextForcedToolResult = checkForForcedToolUsage(
+          currentResponse,
+          nextPayload.tool_choice,
+          forcedTools,
+          usedForcedTools
+        )
+        hasUsedForcedTool = nextForcedToolResult.hasUsedForcedTool
+        usedForcedTools = nextForcedToolResult.usedForcedTools
+        const nextModelEndTime = Date.now()
+        const thisModelTime = nextModelEndTime - nextModelStartTime
+        timeSegments.push({
+          type: 'model',
+          name: request.model,
+          startTime: nextModelStartTime,
+          endTime: nextModelEndTime,
+          duration: thisModelTime,
+        })
+        modelTime += thisModelTime
+        if (currentResponse.choices[0]?.message?.content) {
+          content = currentResponse.choices[0].message.content
+        }
+        if (currentResponse.usage) {
+          tokens.input += currentResponse.usage.prompt_tokens || 0
+          tokens.output += currentResponse.usage.completion_tokens || 0
+          tokens.total += currentResponse.usage.total_tokens || 0
+        }
+        iterationCount++
+      }
+
+      if (iterationCount === MAX_TOOL_ITERATIONS) {
+        enrichLastModelSegmentFromChatCompletions(
+          timeSegments,
+          currentResponse,
+          currentResponse.choices[0]?.message?.tool_calls,
+          { model: request.model, provider: 'fireworks' }
+        )
+      }
+
+      if (request.stream) {
+        const accumulatedCost = calculateCost(requestedModel, tokens.input, tokens.output)
+
+        const streamingParams: ChatCompletionCreateParamsStreaming = {
+          ...payload,
+          messages: [...currentMessages],
+          tool_choice: 'none',
+          stream: true,
+          stream_options: { include_usage: true },
+        }
+
+        if (request.responseFormat) {
+          ;(streamingParams as any).messages = await applyResponseFormat(
+            streamingParams as any,
+            streamingParams.messages,
+            request.responseFormat,
+            requestedModel
+          )
+        }
+
+        const streamResponse = await client.chat.completions.create(
+          streamingParams,
+          request.abortSignal ? { signal: request.abortSignal } : undefined
+        )
+
+        const streamingResult = createStreamingExecution({
+          model: requestedModel,
+          providerStartTime,
+          providerStartTimeISO,
+          timing: {
+            kind: 'accumulated',
+            modelTime,
+            toolsTime,
+            firstResponseTime,
+            iterations: iterationCount + 1,
+            timeSegments,
+          },
+          initialTokens: { input: tokens.input, output: tokens.output, total: tokens.total },
+          initialCost: {
+            input: accumulatedCost.input,
+            output: accumulatedCost.output,
+            total: accumulatedCost.total,
+          },
+          toolCalls:
+            toolCalls.length > 0 ? { list: toolCalls, count: toolCalls.length } : undefined,
+          createStream: ({ output }) =>
+            createReadableStreamFromOpenAIStream(streamResponse, (content, usage) => {
+              output.content = content
+              output.tokens = {
+                input: tokens.input + usage.prompt_tokens,
+                output: tokens.output + usage.completion_tokens,
+                total: tokens.total + usage.total_tokens,
+              }
+
+              const streamCost = calculateCost(
+                requestedModel,
+                usage.prompt_tokens,
+                usage.completion_tokens
+              )
+              const tc = sumToolCosts(toolResults)
+              output.cost = {
+                input: accumulatedCost.input + streamCost.input,
+                output: accumulatedCost.output + streamCost.output,
+                toolCost: tc || undefined,
+                total: accumulatedCost.total + streamCost.total + tc,
+              }
+            }),
+        })
+
+        return streamingResult
+      }
+
+      if (request.responseFormat && hasActiveTools) {
+        const finalPayload: any = {
+          model: payload.model,
+          messages: [...currentMessages],
+        }
+        if (payload.temperature !== undefined) {
+          finalPayload.temperature = payload.temperature
+        }
+        if (payload.max_tokens !== undefined) {
+          finalPayload.max_tokens = payload.max_tokens
+        }
+
+        finalPayload.messages = await applyResponseFormat(
+          finalPayload,
+          finalPayload.messages,
+          request.responseFormat,
+          requestedModel
+        )
+
+        const finalStartTime = Date.now()
+        const finalResponse = await client.chat.completions.create(
+          finalPayload,
+          request.abortSignal ? { signal: request.abortSignal } : undefined
+        )
+        const finalEndTime = Date.now()
+        const finalDuration = finalEndTime - finalStartTime
+
+        timeSegments.push({
+          type: 'model',
+          name: 'Final structured response',
+          startTime: finalStartTime,
+          endTime: finalEndTime,
+          duration: finalDuration,
+        })
+        modelTime += finalDuration
+
+        if (finalResponse.choices[0]?.message?.content) {
+          content = finalResponse.choices[0].message.content
+        }
+        if (finalResponse.usage) {
+          tokens.input += finalResponse.usage.prompt_tokens || 0
+          tokens.output += finalResponse.usage.completion_tokens || 0
+          tokens.total += finalResponse.usage.total_tokens || 0
+        }
+
+        enrichLastModelSegmentFromChatCompletions(
+          timeSegments,
+          finalResponse,
+          finalResponse.choices[0]?.message?.tool_calls,
+          { model: request.model, provider: 'fireworks' }
+        )
+      }
+
+      const providerEndTime = Date.now()
+      const providerEndTimeISO = new Date(providerEndTime).toISOString()
+      const totalDuration = providerEndTime - providerStartTime
+
+      return {
+        content,
+        model: requestedModel,
+        tokens,
+        toolCalls: toolCalls.length > 0 ? toolCalls : undefined,
+        toolResults: toolResults.length > 0 ? toolResults : undefined,
+        timing: {
+          startTime: providerStartTimeISO,
+          endTime: providerEndTimeISO,
+          duration: totalDuration,
+          modelTime: modelTime,
+          toolsTime: toolsTime,
+          firstResponseTime: firstResponseTime,
+          iterations: iterationCount + 1,
+          timeSegments: timeSegments,
+        },
+      }
+    } catch (error) {
+      const providerEndTime = Date.now()
+      const providerEndTimeISO = new Date(providerEndTime).toISOString()
+      const totalDuration = providerEndTime - providerStartTime
+
+      const errorDetails: Record<string, any> = {
+        error: toError(error).message,
+        duration: totalDuration,
+      }
+      if (error && typeof error === 'object') {
+        const err = error as any
+        if (err.status) errorDetails.status = err.status
+        if (err.code) errorDetails.code = err.code
+        if (err.type) errorDetails.type = err.type
+        if (err.error?.message) errorDetails.providerMessage = err.error.message
+        if (err.error?.metadata) errorDetails.metadata = err.error.metadata
+      }
+
+      logger.error('Error in Fireworks request:', errorDetails)
+      throw new ProviderError(toError(error).message, {
+        startTime: providerStartTimeISO,
+        endTime: providerEndTimeISO,
+        duration: totalDuration,
+      })
+    }
+  },
+}

apps/sim/providers/fireworks/utils.ts

@@ -0,0 +1,41 @@
+import type { ChatCompletionChunk } from 'openai/resources/chat/completions'
+import type { CompletionUsage } from 'openai/resources/completions'
+import { checkForForcedToolUsageOpenAI, createOpenAICompatibleStream } from '@/providers/utils'
+
+/**
+ * Checks if a model supports native structured outputs (json_schema).
+ * Fireworks AI supports structured outputs across their inference API.
+ */
+export async function supportsNativeStructuredOutputs(_modelId: string): Promise<boolean> {
+  return true
+}
+
+/**
+ * Creates a ReadableStream from a Fireworks streaming response.
+ * Uses the shared OpenAI-compatible streaming utility.
+ */
+export function createReadableStreamFromOpenAIStream(
+  openaiStream: AsyncIterable<ChatCompletionChunk>,
+  onComplete?: (content: string, usage: CompletionUsage) => void
+): ReadableStream<Uint8Array> {
+  return createOpenAICompatibleStream(openaiStream, 'Fireworks', onComplete)
+}
+
+/**
+ * Checks if a forced tool was used in a Fireworks response.
+ * Uses the shared OpenAI-compatible forced tool usage helper.
+ */
+export function checkForForcedToolUsage(
+  response: any,
+  toolChoice: string | { type: string; function?: { name: string }; name?: string; any?: any },
+  forcedTools: string[],
+  usedForcedTools: string[]
+): { hasUsedForcedTool: boolean; usedForcedTools: string[] } {
+  return checkForForcedToolUsageOpenAI(
+    response,
+    toolChoice,
+    'Fireworks',
+    forcedTools,
+    usedForcedTools
+  )
+}