chore: import upstream snapshot with attribution

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

@@ -0,0 +1,296 @@
+/**
+ * @vitest-environment node
+ */
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+
+const { mockCreate, mockExecuteTool } = vi.hoisted(() => ({
+  mockCreate: vi.fn(),
+  mockExecuteTool: vi.fn(),
+}))
+
+vi.mock('openai', () => ({
+  default: vi.fn().mockImplementation(
+    class {
+      chat = { completions: { create: mockCreate } }
+    }
+  ),
+}))
+
+vi.mock('@/tools', () => ({ executeTool: mockExecuteTool }))
+
+vi.mock('@/providers', () => ({ MAX_TOOL_ITERATIONS: 20 }))
+
+vi.mock('@/lib/core/config/env', () => ({
+  env: { LITELLM_BASE_URL: 'http://litellm.test', LITELLM_API_KEY: '' },
+}))
+
+vi.mock('@/stores/providers', () => ({
+  useProvidersStore: { getState: () => ({ setProviderModels: vi.fn() }) },
+}))
+
+vi.mock('@/providers/models', () => ({
+  getProviderFileAttachment: vi
+    .fn()
+    .mockReturnValue({ maxBytes: 10 * 1024 * 1024, strategy: 'inline' }),
+  INLINE_ATTACHMENT_MAX_BYTES: 10 * 1024 * 1024,
+  getProviderModels: () => [],
+  getProviderDefaultModel: () => '',
+}))
+
+vi.mock('@/providers/attachments', () => ({
+  formatMessagesForProvider: (messages: unknown) => messages,
+}))
+
+vi.mock('@/providers/trace-enrichment', () => ({
+  enrichLastModelSegmentFromChatCompletions: vi.fn(),
+}))
+
+vi.mock('@/providers/litellm/utils', () => ({
+  createReadableStreamFromLiteLLMStream: vi.fn(
+    () => new ReadableStream({ start: (c) => c.close() })
+  ),
+}))
+
+vi.mock('@/providers/utils', () => ({
+  calculateCost: vi.fn(() => ({ input: 0, output: 0, total: 0 })),
+  sumToolCosts: vi.fn(() => 0),
+  prepareToolExecution: vi.fn((_tool, toolArgs) => ({
+    toolParams: toolArgs,
+    executionParams: toolArgs,
+  })),
+  prepareToolsWithUsageControl: vi.fn((tools) => ({
+    tools,
+    toolChoice: 'auto',
+    forcedTools: [],
+    hasFilteredTools: false,
+  })),
+  trackForcedToolUsage: vi.fn(() => ({ hasUsedForcedTool: false, usedForcedTools: [] })),
+  enforceStrictSchema: vi.fn((schema) => ({ ...schema, additionalProperties: false })),
+}))
+
+import { litellmProvider } from '@/providers/litellm'
+import { ProviderError } from '@/providers/types'
+
+interface ChatOptions {
+  content?: string | null
+  toolCalls?: Array<{ id: string; function: { name: string; arguments: string } }>
+  usage?: { prompt_tokens: number; completion_tokens: number; total_tokens: number }
+}
+
+function chat({ content = null, toolCalls, usage }: ChatOptions = {}) {
+  return {
+    choices: [
+      {
+        message: { content, tool_calls: toolCalls },
+        finish_reason: toolCalls ? 'tool_calls' : 'stop',
+      },
+    ],
+    usage: usage ?? { prompt_tokens: 5, completion_tokens: 3, total_tokens: 8 },
+  }
+}
+
+function tool(name: string) {
+  return { id: name, name, description: 'd', parameters: {} }
+}
+
+function run(request: Record<string, unknown>) {
+  return litellmProvider.executeRequest!({
+    model: 'litellm/llama-3',
+    messages: [{ role: 'user', content: 'Hi' }],
+    ...request,
+  } as never) as Promise<any>
+}
+
+const firstPayload = () => mockCreate.mock.calls[0][0]
+const lastPayload = () => mockCreate.mock.calls.at(-1)![0]
+
+describe('litellmProvider.executeRequest', () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+    mockCreate.mockResolvedValue(chat({ content: 'hello' }))
+    mockExecuteTool.mockResolvedValue({ success: true, output: { ok: true } })
+  })
+
+  it('assembles messages, strips the model prefix, and maps params', async () => {
+    const result = await run({
+      systemPrompt: 'You are helpful.',
+      context: 'Some context',
+      temperature: 0.5,
+      maxTokens: 256,
+    })
+
+    const payload = firstPayload()
+    expect(payload.model).toBe('llama-3')
+    expect(payload.messages).toEqual([
+      { role: 'system', content: 'You are helpful.' },
+      { role: 'user', content: 'Some context' },
+      { role: 'user', content: 'Hi' },
+    ])
+    expect(payload.temperature).toBe(0.5)
+    expect(payload.max_completion_tokens).toBe(256)
+    expect(result.content).toBe('hello')
+    expect(result.tokens).toEqual({ input: 5, output: 3, total: 8 })
+  })
+
+  it('forwards reasoning_effort only when set to a non-default value', async () => {
+    await run({ reasoningEffort: 'high' })
+    expect(firstPayload().reasoning_effort).toBe('high')
+
+    mockCreate.mockClear()
+    await run({ reasoningEffort: 'auto' })
+    expect(firstPayload().reasoning_effort).toBeUndefined()
+
+    mockCreate.mockClear()
+    await run({})
+    expect(firstPayload().reasoning_effort).toBeUndefined()
+  })
+
+  it('sanitizes the schema for strict response_format and passes it through otherwise', async () => {
+    await run({ responseFormat: { name: 'r', schema: { type: 'object', properties: {} } } })
+    let rf = firstPayload().response_format
+    expect(rf.type).toBe('json_schema')
+    expect(rf.json_schema.strict).toBe(true)
+    expect(rf.json_schema.schema.additionalProperties).toBe(false)
+
+    mockCreate.mockClear()
+    await run({
+      responseFormat: { name: 'r', schema: { type: 'object', properties: {} }, strict: false },
+    })
+    rf = firstPayload().response_format
+    expect(rf.json_schema.strict).toBe(false)
+    expect(rf.json_schema.schema.additionalProperties).toBeUndefined()
+  })
+
+  it('defers response_format past the tool loop and keeps tools on the final call', async () => {
+    mockCreate
+      .mockResolvedValueOnce(
+        chat({ toolCalls: [{ id: 'c1', function: { name: 'known', arguments: '{"q":1}' } }] })
+      )
+      .mockResolvedValueOnce(chat({ content: 'mid' }))
+      .mockResolvedValueOnce(chat({ content: '{"answer":1}' }))
+
+    const result = await run({
+      tools: [tool('known')],
+      reasoningEffort: 'high',
+      responseFormat: { name: 'r', schema: { type: 'object', properties: {} } },
+    })
+
+    expect(firstPayload().response_format).toBeUndefined()
+    expect(firstPayload().tools).toBeDefined()
+
+    const final = lastPayload()
+    expect(final.response_format.type).toBe('json_schema')
+    expect(final.tools).toBeDefined()
+    expect(final.tool_choice).toBe('none')
+    expect(final.parallel_tool_calls).toBe(false)
+    expect(final.reasoning_effort).toBe('high')
+    expect(result.content).toBe('{"answer":1}')
+  })
+
+  it('defers response_format into the final streaming call while keeping tools', async () => {
+    mockCreate
+      .mockResolvedValueOnce(
+        chat({ toolCalls: [{ id: 'c1', function: { name: 'known', arguments: '{}' } }] })
+      )
+      .mockResolvedValueOnce(chat({ content: 'mid' }))
+
+    const result = await run({
+      stream: true,
+      tools: [tool('known')],
+      responseFormat: { name: 'r', schema: { type: 'object', properties: {} } },
+    })
+
+    const final = lastPayload()
+    expect(final.stream).toBe(true)
+    expect(final.response_format.type).toBe('json_schema')
+    expect(final.tools).toBeDefined()
+    expect(final.tool_choice).toBe('none')
+    expect(final.parallel_tool_calls).toBe(false)
+    expect(result.execution.isStreaming).toBe(true)
+  })
+
+  it('threads assistant tool_calls and a named tool response, and reports toolCalls', async () => {
+    mockCreate
+      .mockResolvedValueOnce(
+        chat({ toolCalls: [{ id: 'c1', function: { name: 'known', arguments: '{}' } }] })
+      )
+      .mockResolvedValueOnce(chat({ content: 'done' }))
+    mockExecuteTool.mockResolvedValue({ success: true, output: { temp: 72 } })
+
+    const result = await run({ tools: [tool('known')] })
+
+    const followupMessages = mockCreate.mock.calls[1][0].messages
+    expect(followupMessages).toContainEqual({
+      role: 'assistant',
+      content: null,
+      tool_calls: [{ id: 'c1', type: 'function', function: { name: 'known', arguments: '{}' } }],
+    })
+    expect(followupMessages).toContainEqual({
+      role: 'tool',
+      tool_call_id: 'c1',
+      name: 'known',
+      content: JSON.stringify({ temp: 72 }),
+    })
+    expect(result.toolCalls).toHaveLength(1)
+    expect(result.content).toBe('done')
+  })
+
+  it('emits a stub tool response for an unanswered tool_call_id', async () => {
+    mockCreate
+      .mockResolvedValueOnce(
+        chat({ toolCalls: [{ id: 'cX', function: { name: 'ghost', arguments: '{}' } }] })
+      )
+      .mockResolvedValueOnce(chat({ content: 'recovered' }))
+
+    await run({ tools: [tool('known')] })
+
+    expect(mockExecuteTool).not.toHaveBeenCalled()
+    const followupMessages = mockCreate.mock.calls[1][0].messages
+    const toolMsg = followupMessages.find((m: any) => m.role === 'tool' && m.tool_call_id === 'cX')
+    expect(toolMsg).toBeDefined()
+    expect(toolMsg.content).toContain('not available')
+  })
+
+  it('executes a tool with empty arguments without failing', async () => {
+    mockCreate
+      .mockResolvedValueOnce(
+        chat({ toolCalls: [{ id: 'c1', function: { name: 'ping', arguments: '' } }] })
+      )
+      .mockResolvedValueOnce(chat({ content: 'pong' }))
+
+    await run({ tools: [tool('ping')] })
+
+    expect(mockExecuteTool).toHaveBeenCalledTimes(1)
+    const toolMsg = mockCreate.mock.calls[1][0].messages.find((m: any) => m.role === 'tool')
+    expect(toolMsg.content).not.toContain('"error":true')
+  })
+
+  it('stops the tool loop at MAX_TOOL_ITERATIONS', async () => {
+    mockCreate.mockResolvedValue(
+      chat({ toolCalls: [{ id: 'c1', function: { name: 'known', arguments: '{}' } }] })
+    )
+
+    await run({ tools: [tool('known')] })
+
+    expect(mockCreate).toHaveBeenCalledTimes(1 + 20)
+    expect(mockExecuteTool).toHaveBeenCalledTimes(20)
+  })
+
+  it('returns a streaming execution when streaming without active tools', async () => {
+    const result = await run({ stream: true })
+
+    expect(firstPayload().stream).toBe(true)
+    expect(firstPayload().stream_options).toEqual({ include_usage: true })
+    expect(result.stream).toBeInstanceOf(ReadableStream)
+    expect(result.execution.isStreaming).toBe(true)
+  })
+
+  it('wraps API errors in a ProviderError using the error envelope message', async () => {
+    mockCreate.mockRejectedValue({
+      error: { message: 'rate limited', type: 'rate_limit_error', code: '429' },
+    })
+
+    await expect(run({})).rejects.toBeInstanceOf(ProviderError)
+    await expect(run({})).rejects.toThrow('rate limited')
+  })
+})

apps/sim/providers/litellm/index.ts

@@ -0,0 +1,722 @@
+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 { env } from '@/lib/core/config/env'
+import type { StreamingExecution } from '@/executor/types'
+import { MAX_TOOL_ITERATIONS } from '@/providers'
+import { formatMessagesForProvider } from '@/providers/attachments'
+import { createReadableStreamFromLiteLLMStream } from '@/providers/litellm/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 {
+  Message,
+  ProviderConfig,
+  ProviderRequest,
+  ProviderResponse,
+  TimeSegment,
+} from '@/providers/types'
+import { ProviderError } from '@/providers/types'
+import {
+  calculateCost,
+  enforceStrictSchema,
+  prepareToolExecution,
+  prepareToolsWithUsageControl,
+  sumToolCosts,
+  trackForcedToolUsage,
+} from '@/providers/utils'
+import { useProvidersStore } from '@/stores/providers'
+import { executeTool } from '@/tools'
+
+const logger = createLogger('LiteLLMProvider')
+const LITELLM_VERSION = '1.0.0'
+
+export const litellmProvider: ProviderConfig = {
+  id: 'litellm',
+  name: 'LiteLLM',
+  description: 'LiteLLM proxy with OpenAI-compatible API',
+  version: LITELLM_VERSION,
+  models: getProviderModels('litellm'),
+  defaultModel: getProviderDefaultModel('litellm'),
+
+  async initialize() {
+    if (typeof window !== 'undefined') {
+      logger.info('Skipping LiteLLM initialization on client side to avoid CORS issues')
+      return
+    }
+
+    const baseUrl = (env.LITELLM_BASE_URL || '').replace(/\/$/, '')
+    if (!baseUrl) {
+      logger.info('LITELLM_BASE_URL not configured, skipping initialization')
+      return
+    }
+
+    try {
+      const headers: Record<string, string> = {
+        'Content-Type': 'application/json',
+      }
+
+      if (env.LITELLM_API_KEY) {
+        headers.Authorization = `Bearer ${env.LITELLM_API_KEY}`
+      }
+
+      const response = await fetch(`${baseUrl}/v1/models`, { headers })
+      if (!response.ok) {
+        await response.text().catch(() => {})
+        useProvidersStore.getState().setProviderModels('litellm', [])
+        logger.warn('LiteLLM service is not available. The provider will be disabled.')
+        return
+      }
+
+      const { vllmUpstreamResponseSchema } = await import('@/lib/api/contracts/providers')
+      const data = vllmUpstreamResponseSchema.parse(await response.json())
+      const models = data.data.map((model) => `litellm/${model.id}`)
+
+      this.models = models
+      useProvidersStore.getState().setProviderModels('litellm', models)
+
+      logger.info(`Discovered ${models.length} LiteLLM model(s):`, { models })
+    } catch (error) {
+      logger.warn('LiteLLM model instantiation failed. The provider will be disabled.', {
+        error: getErrorMessage(error, 'Unknown error'),
+      })
+    }
+  },
+
+  executeRequest: async (
+    request: ProviderRequest
+  ): Promise<ProviderResponse | StreamingExecution> => {
+    logger.info('Preparing LiteLLM request', {
+      model: request.model,
+      hasSystemPrompt: !!request.systemPrompt,
+      hasMessages: !!request.messages?.length,
+      hasTools: !!request.tools?.length,
+      toolCount: request.tools?.length || 0,
+      hasResponseFormat: !!request.responseFormat,
+      stream: !!request.stream,
+    })
+
+    const baseUrl = (env.LITELLM_BASE_URL || '').replace(/\/$/, '')
+    if (!baseUrl) {
+      throw new Error('LITELLM_BASE_URL is required for LiteLLM provider')
+    }
+
+    const apiKey = request.apiKey || env.LITELLM_API_KEY || 'empty'
+    const litellm = new OpenAI({
+      apiKey,
+      baseURL: `${baseUrl}/v1`,
+    })
+
+    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, 'litellm') as Message[]
+
+    const tools = request.tools?.length
+      ? request.tools.map((tool) => adaptOpenAIChatToolSchema(tool))
+      : undefined
+
+    const payload: any = {
+      model: request.model.replace(/^litellm\//, ''),
+      messages: formattedMessages,
+    }
+
+    if (request.temperature !== undefined) payload.temperature = request.temperature
+    if (request.maxTokens != null) payload.max_completion_tokens = request.maxTokens
+
+    if (request.reasoningEffort !== undefined && request.reasoningEffort !== 'auto') {
+      payload.reasoning_effort = request.reasoningEffort
+    }
+
+    const isStrictResponseFormat = request.responseFormat
+      ? request.responseFormat.strict !== false
+      : false
+
+    const responseFormatPayload = request.responseFormat
+      ? {
+          type: 'json_schema' as const,
+          json_schema: {
+            name: request.responseFormat.name || 'response_schema',
+            schema: isStrictResponseFormat
+              ? enforceStrictSchema(request.responseFormat.schema || request.responseFormat)
+              : request.responseFormat.schema || request.responseFormat,
+            strict: isStrictResponseFormat,
+          },
+        }
+      : undefined
+
+    let preparedTools: ReturnType<typeof prepareToolsWithUsageControl> | null = null
+    let hasActiveTools = false
+
+    if (tools?.length) {
+      preparedTools = prepareToolsWithUsageControl(tools, request.tools, logger, 'litellm')
+      const { tools: filteredTools, toolChoice } = preparedTools
+
+      if (filteredTools?.length && toolChoice) {
+        payload.tools = filteredTools
+        payload.tool_choice = toolChoice
+        hasActiveTools = true
+
+        logger.info('LiteLLM request configuration:', {
+          toolCount: filteredTools.length,
+          toolChoice:
+            typeof toolChoice === 'string'
+              ? toolChoice
+              : toolChoice.type === 'function'
+                ? `force:${toolChoice.function.name}`
+                : 'unknown',
+          model: payload.model,
+        })
+      }
+    }
+
+    const deferResponseFormat = !!responseFormatPayload && hasActiveTools
+    if (responseFormatPayload && !deferResponseFormat) {
+      payload.response_format = responseFormatPayload
+      logger.info('Added JSON schema response format to LiteLLM request')
+    }
+
+    const providerStartTime = Date.now()
+    const providerStartTimeISO = new Date(providerStartTime).toISOString()
+
+    try {
+      if (request.stream && (!tools || tools.length === 0 || !hasActiveTools)) {
+        logger.info('Using streaming response for LiteLLM request')
+
+        const streamingParams: ChatCompletionCreateParamsStreaming = {
+          ...payload,
+          stream: true,
+          stream_options: { include_usage: true },
+        }
+        const streamResponse = await litellm.chat.completions.create(
+          streamingParams,
+          request.abortSignal ? { signal: request.abortSignal } : undefined
+        )
+
+        const streamingResult = createStreamingExecution({
+          model: request.model,
+          providerStartTime,
+          providerStartTimeISO,
+          timing: { kind: 'simple', segmentName: request.model },
+          initialTokens: { input: 0, output: 0, total: 0 },
+          initialCost: { input: 0, output: 0, total: 0 },
+          isStreaming: true,
+          createStream: ({ output, finalizeTiming }) =>
+            createReadableStreamFromLiteLLMStream(streamResponse, (content, usage) => {
+              let cleanContent = content
+              if (cleanContent && request.responseFormat) {
+                cleanContent = cleanContent.replace(/```json\n?|\n?```/g, '').trim()
+              }
+
+              output.content = cleanContent
+              output.tokens = {
+                input: usage.prompt_tokens,
+                output: usage.completion_tokens,
+                total: usage.total_tokens,
+              }
+
+              const costResult = calculateCost(
+                request.model,
+                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[] = []
+
+      const checkForForcedToolUsage = (
+        response: any,
+        toolChoice: string | { type: string; function?: { name: string }; name?: string; any?: any }
+      ) => {
+        if (typeof toolChoice === 'object' && response.choices[0]?.message?.tool_calls) {
+          const toolCallsResponse = response.choices[0].message.tool_calls
+          const result = trackForcedToolUsage(
+            toolCallsResponse,
+            toolChoice,
+            logger,
+            'litellm',
+            forcedTools,
+            usedForcedTools
+          )
+          hasUsedForcedTool = result.hasUsedForcedTool
+          usedForcedTools = result.usedForcedTools
+        }
+      }
+
+      let currentResponse = await litellm.chat.completions.create(
+        payload,
+        request.abortSignal ? { signal: request.abortSignal } : undefined
+      )
+      const firstResponseTime = Date.now() - initialCallTime
+
+      let content = currentResponse.choices[0]?.message?.content || ''
+
+      if (content && request.responseFormat) {
+        content = content.replace(/```json\n?|\n?```/g, '').trim()
+      }
+
+      const tokens = {
+        input: currentResponse.usage?.prompt_tokens || 0,
+        output: currentResponse.usage?.completion_tokens || 0,
+        total: currentResponse.usage?.total_tokens || 0,
+      }
+      const toolCalls = []
+      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,
+        },
+      ]
+
+      checkForForcedToolUsage(currentResponse, originalToolChoice)
+
+      while (iterationCount < MAX_TOOL_ITERATIONS) {
+        if (currentResponse.choices[0]?.message?.content) {
+          content = currentResponse.choices[0].message.content
+          if (request.responseFormat) {
+            content = content.replace(/```json\n?|\n?```/g, '').trim()
+          }
+        }
+
+        const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
+
+        enrichLastModelSegmentFromChatCompletions(
+          timeSegments,
+          currentResponse,
+          toolCallsInResponse,
+          { model: request.model, provider: 'litellm' }
+        )
+
+        if (!toolCallsInResponse || toolCallsInResponse.length === 0) {
+          break
+        }
+
+        logger.info(
+          `Processing ${toolCallsInResponse.length} tool calls (iteration ${iterationCount + 1}/${MAX_TOOL_ITERATIONS})`
+        )
+
+        const toolsStartTime = Date.now()
+
+        const toolExecutionPromises = toolCallsInResponse.map(async (toolCall) => {
+          const toolCallStartTime = Date.now()
+          const toolName = toolCall.function.name
+
+          try {
+            const toolArgs = toolCall.function.arguments
+              ? 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:', { error, 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,
+            },
+          })),
+        })
+
+        const respondedToolCallIds = new Set<string>()
+
+        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 && result.output) {
+            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,
+            name: toolName,
+            content: JSON.stringify(resultContent),
+          })
+          respondedToolCallIds.add(toolCall.id)
+        }
+
+        for (const tc of toolCallsInResponse) {
+          if (respondedToolCallIds.has(tc.id)) continue
+          currentMessages.push({
+            role: 'tool',
+            tool_call_id: tc.id,
+            name: tc.function.name,
+            content: JSON.stringify({
+              error: true,
+              message: `Tool "${tc.function.name}" is not available`,
+              tool: tc.function.name,
+            }),
+          })
+        }
+
+        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] },
+            }
+            logger.info(`Forcing next tool: ${remainingTools[0]}`)
+          } else {
+            nextPayload.tool_choice = 'auto'
+            logger.info('All forced tools have been used, switching to auto tool_choice')
+          }
+        }
+
+        const nextModelStartTime = Date.now()
+
+        currentResponse = await litellm.chat.completions.create(
+          nextPayload,
+          request.abortSignal ? { signal: request.abortSignal } : undefined
+        )
+
+        checkForForcedToolUsage(currentResponse, nextPayload.tool_choice)
+
+        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 (request.responseFormat) {
+            content = content.replace(/```json\n?|\n?```/g, '').trim()
+          }
+        }
+
+        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: 'litellm' }
+        )
+      }
+
+      if (request.stream) {
+        logger.info('Using streaming for final response after tool processing')
+
+        const accumulatedCost = calculateCost(request.model, tokens.input, tokens.output)
+
+        const streamingParams: ChatCompletionCreateParamsStreaming = {
+          ...payload,
+          messages: currentMessages,
+          tool_choice: 'none',
+          stream: true,
+          stream_options: { include_usage: true },
+        }
+        if (deferResponseFormat && responseFormatPayload) {
+          streamingParams.response_format = responseFormatPayload
+          streamingParams.parallel_tool_calls = false
+        }
+        const streamResponse = await litellm.chat.completions.create(
+          streamingParams,
+          request.abortSignal ? { signal: request.abortSignal } : undefined
+        )
+
+        const streamingResult = createStreamingExecution({
+          model: request.model,
+          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,
+          isStreaming: true,
+          createStream: ({ output }) =>
+            createReadableStreamFromLiteLLMStream(streamResponse, (content, usage) => {
+              let cleanContent = content
+              if (cleanContent && request.responseFormat) {
+                cleanContent = cleanContent.replace(/```json\n?|\n?```/g, '').trim()
+              }
+
+              output.content = cleanContent
+              output.tokens = {
+                input: tokens.input + usage.prompt_tokens,
+                output: tokens.output + usage.completion_tokens,
+                total: tokens.total + usage.total_tokens,
+              }
+
+              const streamCost = calculateCost(
+                request.model,
+                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 (deferResponseFormat && responseFormatPayload) {
+        logger.info('Applying deferred JSON schema response format after tool processing')
+
+        const finalFormatStartTime = Date.now()
+        const finalPayload: any = {
+          ...payload,
+          messages: currentMessages,
+          response_format: responseFormatPayload,
+          tool_choice: 'none',
+          parallel_tool_calls: false,
+        }
+
+        currentResponse = await litellm.chat.completions.create(
+          finalPayload,
+          request.abortSignal ? { signal: request.abortSignal } : undefined
+        )
+
+        const finalFormatEndTime = Date.now()
+        timeSegments.push({
+          type: 'model',
+          name: request.model,
+          startTime: finalFormatStartTime,
+          endTime: finalFormatEndTime,
+          duration: finalFormatEndTime - finalFormatStartTime,
+        })
+        modelTime += finalFormatEndTime - finalFormatStartTime
+
+        const formattedContent = currentResponse.choices[0]?.message?.content
+        if (formattedContent) {
+          content = formattedContent.replace(/```json\n?|\n?```/g, '').trim()
+        }
+
+        if (currentResponse.usage) {
+          tokens.input += currentResponse.usage.prompt_tokens || 0
+          tokens.output += currentResponse.usage.completion_tokens || 0
+          tokens.total += currentResponse.usage.total_tokens || 0
+        }
+
+        enrichLastModelSegmentFromChatCompletions(
+          timeSegments,
+          currentResponse,
+          currentResponse.choices[0]?.message?.tool_calls,
+          { model: request.model, provider: 'litellm' }
+        )
+      }
+
+      const providerEndTime = Date.now()
+      const providerEndTimeISO = new Date(providerEndTime).toISOString()
+      const totalDuration = providerEndTime - providerStartTime
+
+      return {
+        content,
+        model: request.model,
+        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
+
+      let errorMessage = toError(error).message
+      let errorType: string | undefined
+      let errorCode: string | number | undefined
+
+      if (error && typeof error === 'object' && 'error' in error) {
+        const litellmError = error.error as any
+        if (litellmError && typeof litellmError === 'object') {
+          errorMessage = litellmError.message || errorMessage
+          errorType = litellmError.type
+          errorCode = litellmError.code
+        }
+      }
+
+      logger.error('Error in LiteLLM request:', {
+        error: errorMessage,
+        errorType,
+        errorCode,
+        duration: totalDuration,
+      })
+
+      throw new ProviderError(errorMessage, {
+        startTime: providerStartTimeISO,
+        endTime: providerEndTimeISO,
+        duration: totalDuration,
+      })
+    }
+  },
+}

apps/sim/providers/litellm/utils.ts

@@ -0,0 +1,14 @@
+import type { ChatCompletionChunk } from 'openai/resources/chat/completions'
+import type { CompletionUsage } from 'openai/resources/completions'
+import { createOpenAICompatibleStream } from '@/providers/utils'
+
+/**
+ * Creates a ReadableStream from a LiteLLM streaming response.
+ * Uses the shared OpenAI-compatible streaming utility.
+ */
+export function createReadableStreamFromLiteLLMStream(
+  litellmStream: AsyncIterable<ChatCompletionChunk>,
+  onComplete?: (content: string, usage: CompletionUsage) => void
+): ReadableStream<Uint8Array> {
+  return createOpenAICompatibleStream(litellmStream, 'LiteLLM', onComplete)
+}