mirror of
https://github.com/simstudioai/sim.git
synced 2026-09-24 15:45:35 +08:00
feat(logs): added gantt chart/trace span for workflow executions
This commit is contained in:
Waleed Latif
@@ -257,7 +257,7 @@ export class AgentBlockHandler implements BlockHandler {
|
||||
toolCallsCount: response.toolCalls?.length || 0,
|
||||
})
|
||||
|
||||
// For structured responses, try to parse the content
|
||||
// If structured responses, try to parse the content
|
||||
if (responseFormat) {
|
||||
try {
|
||||
const parsedContent = JSON.parse(response.content)
|
||||
@@ -284,6 +284,7 @@ export class AgentBlockHandler implements BlockHandler {
|
||||
count: response.toolCalls.length,
|
||||
}
|
||||
: undefined,
|
||||
providerTiming: response.timing || undefined,
|
||||
},
|
||||
}
|
||||
|
||||
@@ -316,6 +317,7 @@ export class AgentBlockHandler implements BlockHandler {
|
||||
: [],
|
||||
count: response.toolCalls?.length || 0,
|
||||
},
|
||||
providerTiming: response.timing || undefined,
|
||||
},
|
||||
}
|
||||
}
|
||||
@@ -345,6 +347,7 @@ export class AgentBlockHandler implements BlockHandler {
|
||||
: [],
|
||||
count: response.toolCalls?.length || 0,
|
||||
},
|
||||
providerTiming: response.timing || undefined,
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
@@ -403,6 +403,10 @@ export async function persistExecutionLogs(
|
||||
duration: result.success ? `${totalDuration}ms` : 'NA',
|
||||
trigger: triggerType,
|
||||
createdAt: new Date(),
|
||||
metadata: {
|
||||
traceSpans: (result as any).traceSpans || [],
|
||||
totalDuration: (result as any).totalDuration || totalDuration,
|
||||
},
|
||||
})
|
||||
} catch (error: any) {
|
||||
logger.error(`Error persisting execution logs: ${error.message}`, { error })
|
||||
|
||||
@@ -1,7 +1,7 @@
|
||||
import Anthropic from '@anthropic-ai/sdk'
|
||||
import { createLogger } from '@/lib/logs/console-logger'
|
||||
import { executeTool } from '@/tools'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse } from '../types'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse, TimeSegment } from '../types'
|
||||
|
||||
const logger = createLogger('Anthropic Provider')
|
||||
|
||||
@@ -186,153 +186,251 @@ ${fieldDescriptions}
|
||||
payload.tools = tools
|
||||
}
|
||||
|
||||
// Make the initial API request
|
||||
let currentResponse = await anthropic.messages.create(payload)
|
||||
let content = ''
|
||||
|
||||
// Extract text content from the message
|
||||
if (Array.isArray(currentResponse.content)) {
|
||||
content = currentResponse.content
|
||||
.filter((item) => item.type === 'text')
|
||||
.map((item) => item.text)
|
||||
.join('\n')
|
||||
}
|
||||
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.input_tokens || 0,
|
||||
completion: currentResponse.usage?.output_tokens || 0,
|
||||
total:
|
||||
(currentResponse.usage?.input_tokens || 0) + (currentResponse.usage?.output_tokens || 0),
|
||||
}
|
||||
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...messages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
// Start execution timer for the entire provider execution
|
||||
const providerStartTime = Date.now()
|
||||
const providerStartTimeISO = new Date(providerStartTime).toISOString()
|
||||
|
||||
try {
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
// Check for tool calls
|
||||
const toolUses = currentResponse.content.filter((item) => item.type === 'tool_use')
|
||||
if (!toolUses || toolUses.length === 0) {
|
||||
break
|
||||
}
|
||||
// Make the initial API request
|
||||
const initialCallTime = Date.now()
|
||||
let currentResponse = await anthropic.messages.create(payload)
|
||||
const firstResponseTime = Date.now() - initialCallTime
|
||||
|
||||
// Process each tool call
|
||||
for (const toolUse of toolUses) {
|
||||
try {
|
||||
const toolName = toolUse.name
|
||||
const toolArgs = toolUse.input as Record<string, any>
|
||||
let content = ''
|
||||
|
||||
// Get the tool from the tools registry
|
||||
const tool = request.tools?.find((t) => t.id === toolName)
|
||||
if (!tool) continue
|
||||
|
||||
// Execute the tool
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
|
||||
if (!result.success) continue
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: result.timing?.startTime,
|
||||
endTime: result.timing?.endTime,
|
||||
duration: result.timing?.duration,
|
||||
result: result.output,
|
||||
})
|
||||
|
||||
// Add the tool call and result to messages
|
||||
const toolUseId = generateToolUseId(toolName)
|
||||
|
||||
currentMessages.push({
|
||||
role: 'assistant',
|
||||
content: [
|
||||
{
|
||||
type: 'tool_use',
|
||||
id: toolUseId,
|
||||
name: toolName,
|
||||
input: toolArgs,
|
||||
} as any,
|
||||
],
|
||||
})
|
||||
|
||||
currentMessages.push({
|
||||
role: 'user',
|
||||
content: [
|
||||
{
|
||||
type: 'tool_result',
|
||||
tool_use_id: toolUseId,
|
||||
content: JSON.stringify(result.output),
|
||||
} as any,
|
||||
],
|
||||
})
|
||||
} catch (error) {
|
||||
logger.error('Error processing tool call:', { error })
|
||||
}
|
||||
}
|
||||
|
||||
// Make the next request with updated messages
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
|
||||
// Make the next request
|
||||
currentResponse = await anthropic.messages.create(nextPayload)
|
||||
|
||||
// Update content if we have a text response
|
||||
const textContent = currentResponse.content
|
||||
// Extract text content from the message
|
||||
if (Array.isArray(currentResponse.content)) {
|
||||
content = currentResponse.content
|
||||
.filter((item) => item.type === 'text')
|
||||
.map((item) => item.text)
|
||||
.join('\n')
|
||||
}
|
||||
|
||||
if (textContent) {
|
||||
content = textContent
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.input_tokens || 0,
|
||||
completion: currentResponse.usage?.output_tokens || 0,
|
||||
total:
|
||||
(currentResponse.usage?.input_tokens || 0) + (currentResponse.usage?.output_tokens || 0),
|
||||
}
|
||||
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...messages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
|
||||
// Track time spent in model vs tools
|
||||
let modelTime = firstResponseTime
|
||||
let toolsTime = 0
|
||||
|
||||
// Track each model and tool call segment with timestamps
|
||||
const timeSegments: TimeSegment[] = [
|
||||
{
|
||||
type: 'model',
|
||||
name: 'Initial response',
|
||||
startTime: initialCallTime,
|
||||
endTime: initialCallTime + firstResponseTime,
|
||||
duration: firstResponseTime,
|
||||
},
|
||||
]
|
||||
|
||||
try {
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
// Check for tool calls
|
||||
const toolUses = currentResponse.content.filter((item) => item.type === 'tool_use')
|
||||
if (!toolUses || toolUses.length === 0) {
|
||||
break
|
||||
}
|
||||
|
||||
// Track time for tool calls in this batch
|
||||
const toolsStartTime = Date.now()
|
||||
|
||||
// Process each tool call
|
||||
for (const toolUse of toolUses) {
|
||||
try {
|
||||
const toolName = toolUse.name
|
||||
const toolArgs = toolUse.input as Record<string, any>
|
||||
|
||||
// Get the tool from the tools registry
|
||||
const tool = request.tools?.find((t) => t.id === toolName)
|
||||
if (!tool) continue
|
||||
|
||||
// Execute the tool
|
||||
const toolCallStartTime = Date.now()
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
const toolCallEndTime = Date.now()
|
||||
const toolCallDuration = toolCallEndTime - toolCallStartTime
|
||||
|
||||
if (!result.success) continue
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'tool',
|
||||
name: toolName,
|
||||
startTime: toolCallStartTime,
|
||||
endTime: toolCallEndTime,
|
||||
duration: toolCallDuration,
|
||||
})
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: new Date(toolCallStartTime).toISOString(),
|
||||
endTime: new Date(toolCallEndTime).toISOString(),
|
||||
duration: toolCallDuration,
|
||||
result: result.output,
|
||||
})
|
||||
|
||||
// Add the tool call and result to messages
|
||||
const toolUseId = generateToolUseId(toolName)
|
||||
|
||||
currentMessages.push({
|
||||
role: 'assistant',
|
||||
content: [
|
||||
{
|
||||
type: 'tool_use',
|
||||
id: toolUseId,
|
||||
name: toolName,
|
||||
input: toolArgs,
|
||||
} as any,
|
||||
],
|
||||
})
|
||||
|
||||
currentMessages.push({
|
||||
role: 'user',
|
||||
content: [
|
||||
{
|
||||
type: 'tool_result',
|
||||
tool_use_id: toolUseId,
|
||||
content: JSON.stringify(result.output),
|
||||
} as any,
|
||||
],
|
||||
})
|
||||
} catch (error) {
|
||||
logger.error('Error processing tool call:', { error })
|
||||
}
|
||||
}
|
||||
|
||||
// Calculate tool call time for this iteration
|
||||
const thisToolsTime = Date.now() - toolsStartTime
|
||||
toolsTime += thisToolsTime
|
||||
|
||||
// Make the next request with updated messages
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
|
||||
// Time the next model call
|
||||
const nextModelStartTime = Date.now()
|
||||
|
||||
// Make the next request
|
||||
currentResponse = await anthropic.messages.create(nextPayload)
|
||||
|
||||
const nextModelEndTime = Date.now()
|
||||
const thisModelTime = nextModelEndTime - nextModelStartTime
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'model',
|
||||
name: `Model response (iteration ${iterationCount + 1})`,
|
||||
startTime: nextModelStartTime,
|
||||
endTime: nextModelEndTime,
|
||||
duration: thisModelTime,
|
||||
})
|
||||
|
||||
// Add to model time
|
||||
modelTime += thisModelTime
|
||||
|
||||
// Update content if we have a text response
|
||||
const textContent = currentResponse.content
|
||||
.filter((item) => item.type === 'text')
|
||||
.map((item) => item.text)
|
||||
.join('\n')
|
||||
|
||||
if (textContent) {
|
||||
content = textContent
|
||||
}
|
||||
|
||||
// Update token counts
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.input_tokens || 0
|
||||
tokens.completion += currentResponse.usage.output_tokens || 0
|
||||
tokens.total +=
|
||||
(currentResponse.usage.input_tokens || 0) + (currentResponse.usage.output_tokens || 0)
|
||||
}
|
||||
|
||||
iterationCount++
|
||||
}
|
||||
} catch (error) {
|
||||
logger.error('Error in Anthropic request:', { error })
|
||||
throw error
|
||||
}
|
||||
|
||||
// Update token counts
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.input_tokens || 0
|
||||
tokens.completion += currentResponse.usage.output_tokens || 0
|
||||
tokens.total +=
|
||||
(currentResponse.usage.input_tokens || 0) + (currentResponse.usage.output_tokens || 0)
|
||||
// If the content looks like it contains JSON, extract just the JSON part
|
||||
if (content.includes('{') && content.includes('}')) {
|
||||
try {
|
||||
const jsonMatch = content.match(/\{[\s\S]*\}/m)
|
||||
if (jsonMatch) {
|
||||
content = jsonMatch[0]
|
||||
}
|
||||
} catch (e) {
|
||||
logger.error('Error extracting JSON from response:', { error: e })
|
||||
}
|
||||
}
|
||||
|
||||
iterationCount++
|
||||
// Calculate overall timing
|
||||
const providerEndTime = Date.now()
|
||||
const providerEndTimeISO = new Date(providerEndTime).toISOString()
|
||||
const totalDuration = providerEndTime - providerStartTime
|
||||
|
||||
return {
|
||||
content,
|
||||
model: request.model || 'claude-3-7-sonnet-20250219',
|
||||
tokens,
|
||||
toolCalls:
|
||||
toolCalls.length > 0
|
||||
? toolCalls.map((tc) => ({
|
||||
name: tc.name,
|
||||
arguments: tc.arguments as Record<string, any>,
|
||||
startTime: tc.startTime,
|
||||
endTime: tc.endTime,
|
||||
duration: tc.duration,
|
||||
result: tc.result,
|
||||
}))
|
||||
: 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) {
|
||||
logger.error('Error in Anthropic request:', { error })
|
||||
throw error
|
||||
}
|
||||
// Include timing information even for errors
|
||||
const providerEndTime = Date.now()
|
||||
const providerEndTimeISO = new Date(providerEndTime).toISOString()
|
||||
const totalDuration = providerEndTime - providerStartTime
|
||||
|
||||
// If the content looks like it contains JSON, extract just the JSON part
|
||||
if (content.includes('{') && content.includes('}')) {
|
||||
try {
|
||||
const jsonMatch = content.match(/\{[\s\S]*\}/m)
|
||||
if (jsonMatch) {
|
||||
content = jsonMatch[0]
|
||||
}
|
||||
} catch (e) {
|
||||
logger.error('Error extracting JSON from response:', { error: e })
|
||||
logger.error('Error in Anthropic request:', { error, duration: totalDuration })
|
||||
|
||||
// Create a new error with timing information
|
||||
const enhancedError = new Error(error instanceof Error ? error.message : String(error))
|
||||
// @ts-ignore - Adding timing property to the error
|
||||
enhancedError.timing = {
|
||||
startTime: providerStartTimeISO,
|
||||
endTime: providerEndTimeISO,
|
||||
duration: totalDuration,
|
||||
}
|
||||
}
|
||||
|
||||
return {
|
||||
content,
|
||||
model: request.model || 'claude-3-7-sonnet-20250219',
|
||||
tokens,
|
||||
toolCalls:
|
||||
toolCalls.length > 0
|
||||
? toolCalls.map((tc) => ({
|
||||
name: tc.name,
|
||||
arguments: tc.arguments as Record<string, any>,
|
||||
}))
|
||||
: undefined,
|
||||
toolResults: toolResults.length > 0 ? toolResults : undefined,
|
||||
throw enhancedError
|
||||
}
|
||||
},
|
||||
}
|
||||
|
||||
@@ -1,7 +1,7 @@
|
||||
import { Cerebras } from '@cerebras/cerebras_cloud_sdk'
|
||||
import { createLogger } from '@/lib/logs/console-logger'
|
||||
import { executeTool } from '@/tools'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse } from '../types'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse, TimeSegment } from '../types'
|
||||
|
||||
const logger = createLogger('Cerebras Provider')
|
||||
|
||||
@@ -17,6 +17,10 @@ export const cerebrasProvider: ProviderConfig = {
|
||||
throw new Error('API key is required for Cerebras')
|
||||
}
|
||||
|
||||
// Start execution timer for the entire provider execution
|
||||
const providerStartTime = Date.now()
|
||||
const providerStartTimeISO = new Date(providerStartTime).toISOString()
|
||||
|
||||
try {
|
||||
const client = new Cerebras({
|
||||
apiKey: request.apiKey,
|
||||
@@ -83,7 +87,9 @@ export const cerebrasProvider: ProviderConfig = {
|
||||
}
|
||||
|
||||
// Make the initial API request
|
||||
const initialCallTime = Date.now()
|
||||
let currentResponse = (await client.chat.completions.create(payload)) as CerebrasResponse
|
||||
const firstResponseTime = Date.now() - initialCallTime
|
||||
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
let tokens = {
|
||||
@@ -97,6 +103,21 @@ export const cerebrasProvider: ProviderConfig = {
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
|
||||
// Track time spent in model vs tools
|
||||
let modelTime = firstResponseTime
|
||||
let toolsTime = 0
|
||||
|
||||
// Track each model and tool call segment with timestamps
|
||||
const timeSegments: TimeSegment[] = [
|
||||
{
|
||||
type: 'model',
|
||||
name: 'Initial response',
|
||||
startTime: initialCallTime,
|
||||
endTime: initialCallTime + firstResponseTime,
|
||||
duration: firstResponseTime,
|
||||
},
|
||||
]
|
||||
|
||||
// Keep track of processed tool calls to avoid duplicates
|
||||
const processedToolCallIds = new Set()
|
||||
// Keep track of tool call signatures to detect repeats
|
||||
@@ -115,6 +136,9 @@ export const cerebrasProvider: ProviderConfig = {
|
||||
break
|
||||
}
|
||||
|
||||
// Track time for tool calls in this batch
|
||||
const toolsStartTime = Date.now()
|
||||
|
||||
// Process each tool call
|
||||
let processedAnyToolCall = false
|
||||
let hasRepeatedToolCalls = false
|
||||
@@ -145,18 +169,30 @@ export const cerebrasProvider: ProviderConfig = {
|
||||
if (!tool) continue
|
||||
|
||||
// Execute the tool
|
||||
const toolCallStartTime = Date.now()
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
const toolCallEndTime = Date.now()
|
||||
const toolCallDuration = toolCallEndTime - toolCallStartTime
|
||||
|
||||
if (!result.success) continue
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'tool',
|
||||
name: toolName,
|
||||
startTime: toolCallStartTime,
|
||||
endTime: toolCallEndTime,
|
||||
duration: toolCallDuration,
|
||||
})
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: result.timing?.startTime,
|
||||
endTime: result.timing?.endTime,
|
||||
duration: result.timing?.duration,
|
||||
startTime: new Date(toolCallStartTime).toISOString(),
|
||||
endTime: new Date(toolCallEndTime).toISOString(),
|
||||
duration: toolCallDuration,
|
||||
result: result.output,
|
||||
})
|
||||
|
||||
@@ -186,8 +222,15 @@ export const cerebrasProvider: ProviderConfig = {
|
||||
}
|
||||
}
|
||||
|
||||
// Calculate tool call time for this iteration
|
||||
const thisToolsTime = Date.now() - toolsStartTime
|
||||
toolsTime += thisToolsTime
|
||||
|
||||
// After processing tool calls, get a final response
|
||||
if (processedAnyToolCall || hasRepeatedToolCalls) {
|
||||
// Time the next model call
|
||||
const nextModelStartTime = Date.now()
|
||||
|
||||
// Make the final request
|
||||
const finalPayload = {
|
||||
...payload,
|
||||
@@ -199,6 +242,21 @@ export const cerebrasProvider: ProviderConfig = {
|
||||
finalPayload
|
||||
)) as CerebrasResponse
|
||||
|
||||
const nextModelEndTime = Date.now()
|
||||
const thisModelTime = nextModelEndTime - nextModelStartTime
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'model',
|
||||
name: `Final response`,
|
||||
startTime: nextModelStartTime,
|
||||
endTime: nextModelEndTime,
|
||||
duration: thisModelTime,
|
||||
})
|
||||
|
||||
// Add to model time
|
||||
modelTime += thisModelTime
|
||||
|
||||
if (finalResponse.choices[0]?.message?.content) {
|
||||
content = finalResponse.choices[0].message.content
|
||||
}
|
||||
@@ -221,11 +279,29 @@ export const cerebrasProvider: ProviderConfig = {
|
||||
messages: currentMessages,
|
||||
}
|
||||
|
||||
// Time the next model call
|
||||
const nextModelStartTime = Date.now()
|
||||
|
||||
// Make the next request
|
||||
currentResponse = (await client.chat.completions.create(
|
||||
nextPayload
|
||||
)) as CerebrasResponse
|
||||
|
||||
const nextModelEndTime = Date.now()
|
||||
const thisModelTime = nextModelEndTime - nextModelStartTime
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'model',
|
||||
name: `Model response (iteration ${iterationCount + 1})`,
|
||||
startTime: nextModelStartTime,
|
||||
endTime: nextModelEndTime,
|
||||
duration: thisModelTime,
|
||||
})
|
||||
|
||||
// Add to model time
|
||||
modelTime += thisModelTime
|
||||
|
||||
// Update token counts
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.prompt_tokens || 0
|
||||
@@ -240,16 +316,46 @@ export const cerebrasProvider: ProviderConfig = {
|
||||
logger.error('Error in Cerebras tool processing:', { error })
|
||||
}
|
||||
|
||||
// Calculate overall timing
|
||||
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) {
|
||||
logger.error('Error in Cerebras request:', { error })
|
||||
throw error
|
||||
// Include timing information even for errors
|
||||
const providerEndTime = Date.now()
|
||||
const providerEndTimeISO = new Date(providerEndTime).toISOString()
|
||||
const totalDuration = providerEndTime - providerStartTime
|
||||
|
||||
logger.error('Error in Cerebras request:', { error, duration: totalDuration })
|
||||
|
||||
// Create a new error with timing information
|
||||
const enhancedError = new Error(error instanceof Error ? error.message : String(error))
|
||||
// @ts-ignore - Adding timing property to the error
|
||||
enhancedError.timing = {
|
||||
startTime: providerStartTimeISO,
|
||||
endTime: providerEndTimeISO,
|
||||
duration: totalDuration,
|
||||
}
|
||||
|
||||
throw enhancedError
|
||||
}
|
||||
},
|
||||
}
|
||||
|
||||
+256
-163
@@ -1,7 +1,7 @@
|
||||
import OpenAI from 'openai'
|
||||
import { createLogger } from '@/lib/logs/console-logger'
|
||||
import { executeTool } from '@/tools'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse } from '../types'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse, TimeSegment } from '../types'
|
||||
|
||||
const logger = createLogger('Deepseek Provider')
|
||||
|
||||
@@ -18,184 +18,277 @@ export const deepseekProvider: ProviderConfig = {
|
||||
throw new Error('API key is required for Deepseek')
|
||||
}
|
||||
|
||||
// Deepseek uses the OpenAI SDK with a custom baseURL
|
||||
const deepseek = new OpenAI({
|
||||
apiKey: request.apiKey,
|
||||
baseURL: 'https://api.deepseek.com/v1',
|
||||
dangerouslyAllowBrowser: true,
|
||||
})
|
||||
|
||||
// Start with an empty array for all messages
|
||||
const allMessages = []
|
||||
|
||||
// Add system prompt if present
|
||||
if (request.systemPrompt) {
|
||||
allMessages.push({
|
||||
role: 'system',
|
||||
content: request.systemPrompt,
|
||||
})
|
||||
}
|
||||
|
||||
// Add context if present
|
||||
if (request.context) {
|
||||
allMessages.push({
|
||||
role: 'user',
|
||||
content: request.context,
|
||||
})
|
||||
}
|
||||
|
||||
// Add remaining messages
|
||||
if (request.messages) {
|
||||
allMessages.push(...request.messages)
|
||||
}
|
||||
|
||||
// Transform tools to OpenAI format if provided
|
||||
const tools = request.tools?.length
|
||||
? request.tools.map((tool) => ({
|
||||
type: 'function',
|
||||
function: {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
parameters: tool.parameters,
|
||||
},
|
||||
}))
|
||||
: undefined
|
||||
|
||||
const payload: any = {
|
||||
model: 'deepseek-chat', // Hardcode to deepseek-chat regardless of what's selected in the UI
|
||||
messages: allMessages,
|
||||
}
|
||||
|
||||
// Add optional parameters
|
||||
if (request.temperature !== undefined) payload.temperature = request.temperature
|
||||
if (request.maxTokens !== undefined) payload.max_tokens = request.maxTokens
|
||||
|
||||
// Add tools if provided
|
||||
if (tools?.length) {
|
||||
payload.tools = tools
|
||||
payload.tool_choice = 'auto'
|
||||
}
|
||||
|
||||
// Make the initial API request
|
||||
let currentResponse = await deepseek.chat.completions.create(payload)
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
|
||||
// Clean up the response content if it exists
|
||||
if (content) {
|
||||
// Remove any markdown code block markers
|
||||
content = content.replace(/```json\n?|\n?```/g, '')
|
||||
// Trim any whitespace
|
||||
content = content.trim()
|
||||
}
|
||||
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.prompt_tokens || 0,
|
||||
completion: currentResponse.usage?.completion_tokens || 0,
|
||||
total: currentResponse.usage?.total_tokens || 0,
|
||||
}
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...allMessages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
// Start execution timer for the entire provider execution
|
||||
const providerStartTime = Date.now()
|
||||
const providerStartTimeISO = new Date(providerStartTime).toISOString()
|
||||
|
||||
try {
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
// Check for tool calls
|
||||
const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
|
||||
if (!toolCallsInResponse || toolCallsInResponse.length === 0) {
|
||||
break
|
||||
}
|
||||
// Deepseek uses the OpenAI SDK with a custom baseURL
|
||||
const deepseek = new OpenAI({
|
||||
apiKey: request.apiKey,
|
||||
baseURL: 'https://api.deepseek.com/v1',
|
||||
dangerouslyAllowBrowser: true,
|
||||
})
|
||||
|
||||
// Process each tool call
|
||||
for (const toolCall of toolCallsInResponse) {
|
||||
try {
|
||||
const toolName = toolCall.function.name
|
||||
const toolArgs = JSON.parse(toolCall.function.arguments)
|
||||
// Start with an empty array for all messages
|
||||
const allMessages = []
|
||||
|
||||
// Get the tool from the tools registry
|
||||
const tool = request.tools?.find((t) => t.id === toolName)
|
||||
if (!tool) continue
|
||||
// Add system prompt if present
|
||||
if (request.systemPrompt) {
|
||||
allMessages.push({
|
||||
role: 'system',
|
||||
content: request.systemPrompt,
|
||||
})
|
||||
}
|
||||
|
||||
// Execute the tool
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
// Add context if present
|
||||
if (request.context) {
|
||||
allMessages.push({
|
||||
role: 'user',
|
||||
content: request.context,
|
||||
})
|
||||
}
|
||||
|
||||
if (!result.success) continue
|
||||
// Add remaining messages
|
||||
if (request.messages) {
|
||||
allMessages.push(...request.messages)
|
||||
}
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: result.timing?.startTime,
|
||||
endTime: result.timing?.endTime,
|
||||
duration: result.timing?.duration,
|
||||
result: result.output,
|
||||
})
|
||||
// Transform tools to OpenAI format if provided
|
||||
const tools = request.tools?.length
|
||||
? request.tools.map((tool) => ({
|
||||
type: 'function',
|
||||
function: {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
parameters: tool.parameters,
|
||||
},
|
||||
}))
|
||||
: undefined
|
||||
|
||||
// Add the tool call and result to messages
|
||||
currentMessages.push({
|
||||
role: 'assistant',
|
||||
content: null,
|
||||
tool_calls: [
|
||||
{
|
||||
id: toolCall.id,
|
||||
type: 'function',
|
||||
function: {
|
||||
name: toolName,
|
||||
arguments: toolCall.function.arguments,
|
||||
},
|
||||
},
|
||||
],
|
||||
})
|
||||
const payload: any = {
|
||||
model: 'deepseek-chat', // Hardcode to deepseek-chat regardless of what's selected in the UI
|
||||
messages: allMessages,
|
||||
}
|
||||
|
||||
currentMessages.push({
|
||||
role: 'tool',
|
||||
tool_call_id: toolCall.id,
|
||||
content: JSON.stringify(result.output),
|
||||
})
|
||||
} catch (error) {
|
||||
logger.error('Error processing tool call:', { error })
|
||||
// Add optional parameters
|
||||
if (request.temperature !== undefined) payload.temperature = request.temperature
|
||||
if (request.maxTokens !== undefined) payload.max_tokens = request.maxTokens
|
||||
|
||||
// Add tools if provided
|
||||
if (tools?.length) {
|
||||
payload.tools = tools
|
||||
payload.tool_choice = 'auto'
|
||||
}
|
||||
|
||||
// Make the initial API request
|
||||
const initialCallTime = Date.now()
|
||||
let currentResponse = await deepseek.chat.completions.create(payload)
|
||||
const firstResponseTime = Date.now() - initialCallTime
|
||||
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
|
||||
// Clean up the response content if it exists
|
||||
if (content) {
|
||||
// Remove any markdown code block markers
|
||||
content = content.replace(/```json\n?|\n?```/g, '')
|
||||
// Trim any whitespace
|
||||
content = content.trim()
|
||||
}
|
||||
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.prompt_tokens || 0,
|
||||
completion: currentResponse.usage?.completion_tokens || 0,
|
||||
total: currentResponse.usage?.total_tokens || 0,
|
||||
}
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...allMessages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
|
||||
// Track time spent in model vs tools
|
||||
let modelTime = firstResponseTime
|
||||
let toolsTime = 0
|
||||
|
||||
// Track each model and tool call segment with timestamps
|
||||
const timeSegments: TimeSegment[] = [
|
||||
{
|
||||
type: 'model',
|
||||
name: 'Initial response',
|
||||
startTime: initialCallTime,
|
||||
endTime: initialCallTime + firstResponseTime,
|
||||
duration: firstResponseTime,
|
||||
},
|
||||
]
|
||||
|
||||
try {
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
// Check for tool calls
|
||||
const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
|
||||
if (!toolCallsInResponse || toolCallsInResponse.length === 0) {
|
||||
break
|
||||
}
|
||||
|
||||
// Track time for tool calls in this batch
|
||||
const toolsStartTime = Date.now()
|
||||
|
||||
// Process each tool call
|
||||
for (const toolCall of toolCallsInResponse) {
|
||||
try {
|
||||
const toolName = toolCall.function.name
|
||||
const toolArgs = JSON.parse(toolCall.function.arguments)
|
||||
|
||||
// Get the tool from the tools registry
|
||||
const tool = request.tools?.find((t) => t.id === toolName)
|
||||
if (!tool) continue
|
||||
|
||||
// Execute the tool
|
||||
const toolCallStartTime = Date.now()
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
const toolCallEndTime = Date.now()
|
||||
const toolCallDuration = toolCallEndTime - toolCallStartTime
|
||||
|
||||
if (!result.success) continue
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'tool',
|
||||
name: toolName,
|
||||
startTime: toolCallStartTime,
|
||||
endTime: toolCallEndTime,
|
||||
duration: toolCallDuration,
|
||||
})
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: new Date(toolCallStartTime).toISOString(),
|
||||
endTime: new Date(toolCallEndTime).toISOString(),
|
||||
duration: toolCallDuration,
|
||||
result: result.output,
|
||||
})
|
||||
|
||||
// Add the tool call and result to messages
|
||||
currentMessages.push({
|
||||
role: 'assistant',
|
||||
content: null,
|
||||
tool_calls: [
|
||||
{
|
||||
id: toolCall.id,
|
||||
type: 'function',
|
||||
function: {
|
||||
name: toolName,
|
||||
arguments: toolCall.function.arguments,
|
||||
},
|
||||
},
|
||||
],
|
||||
})
|
||||
|
||||
currentMessages.push({
|
||||
role: 'tool',
|
||||
tool_call_id: toolCall.id,
|
||||
content: JSON.stringify(result.output),
|
||||
})
|
||||
} catch (error) {
|
||||
logger.error('Error processing tool call:', { error })
|
||||
}
|
||||
}
|
||||
|
||||
// Calculate tool call time for this iteration
|
||||
const thisToolsTime = Date.now() - toolsStartTime
|
||||
toolsTime += thisToolsTime
|
||||
|
||||
// Make the next request with updated messages
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
|
||||
// Time the next model call
|
||||
const nextModelStartTime = Date.now()
|
||||
|
||||
// Make the next request
|
||||
currentResponse = await deepseek.chat.completions.create(nextPayload)
|
||||
|
||||
const nextModelEndTime = Date.now()
|
||||
const thisModelTime = nextModelEndTime - nextModelStartTime
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'model',
|
||||
name: `Model response (iteration ${iterationCount + 1})`,
|
||||
startTime: nextModelStartTime,
|
||||
endTime: nextModelEndTime,
|
||||
duration: thisModelTime,
|
||||
})
|
||||
|
||||
// Add to model time
|
||||
modelTime += thisModelTime
|
||||
|
||||
// Update content if we have a text response
|
||||
if (currentResponse.choices[0]?.message?.content) {
|
||||
content = currentResponse.choices[0].message.content
|
||||
// Clean up the response content
|
||||
content = content.replace(/```json\n?|\n?```/g, '')
|
||||
content = content.trim()
|
||||
}
|
||||
|
||||
// Update token counts
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.prompt_tokens || 0
|
||||
tokens.completion += currentResponse.usage.completion_tokens || 0
|
||||
tokens.total += currentResponse.usage.total_tokens || 0
|
||||
}
|
||||
|
||||
iterationCount++
|
||||
}
|
||||
} catch (error) {
|
||||
logger.error('Error in Deepseek request:', { error })
|
||||
}
|
||||
|
||||
// Make the next request with updated messages
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
// Calculate overall timing
|
||||
const providerEndTime = Date.now()
|
||||
const providerEndTimeISO = new Date(providerEndTime).toISOString()
|
||||
const totalDuration = providerEndTime - providerStartTime
|
||||
|
||||
// Make the next request
|
||||
currentResponse = await deepseek.chat.completions.create(nextPayload)
|
||||
|
||||
// Update content if we have a text response
|
||||
if (currentResponse.choices[0]?.message?.content) {
|
||||
content = currentResponse.choices[0].message.content
|
||||
// Clean up the response content
|
||||
content = content.replace(/```json\n?|\n?```/g, '')
|
||||
content = content.trim()
|
||||
}
|
||||
|
||||
// Update token counts
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.prompt_tokens || 0
|
||||
tokens.completion += currentResponse.usage.completion_tokens || 0
|
||||
tokens.total += currentResponse.usage.total_tokens || 0
|
||||
}
|
||||
|
||||
iterationCount++
|
||||
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) {
|
||||
logger.error('Error in Deepseek request:', { error })
|
||||
throw error
|
||||
}
|
||||
// Include timing information even for errors
|
||||
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,
|
||||
logger.error('Error in Deepseek request:', { error, duration: totalDuration })
|
||||
|
||||
// Create a new error with timing information
|
||||
const enhancedError = new Error(error instanceof Error ? error.message : String(error))
|
||||
// @ts-ignore - Adding timing property to the error
|
||||
enhancedError.timing = {
|
||||
startTime: providerStartTimeISO,
|
||||
endTime: providerEndTimeISO,
|
||||
duration: totalDuration,
|
||||
}
|
||||
|
||||
throw enhancedError
|
||||
}
|
||||
},
|
||||
}
|
||||
|
||||
+254
-161
@@ -1,7 +1,7 @@
|
||||
import OpenAI from 'openai'
|
||||
import { createLogger } from '@/lib/logs/console-logger'
|
||||
import { executeTool } from '@/tools'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse } from '../types'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse, TimeSegment } from '../types'
|
||||
|
||||
const logger = createLogger('Google Provider')
|
||||
|
||||
@@ -18,182 +18,275 @@ export const googleProvider: ProviderConfig = {
|
||||
throw new Error('API key is required for Google Gemini')
|
||||
}
|
||||
|
||||
const openai = new OpenAI({
|
||||
apiKey: request.apiKey,
|
||||
baseURL: 'https://generativelanguage.googleapis.com/v1beta/openai/',
|
||||
dangerouslyAllowBrowser: true,
|
||||
})
|
||||
|
||||
// Start with an empty array for all messages
|
||||
const allMessages = []
|
||||
|
||||
// Add system prompt if present
|
||||
if (request.systemPrompt) {
|
||||
allMessages.push({
|
||||
role: 'system',
|
||||
content: request.systemPrompt,
|
||||
})
|
||||
}
|
||||
|
||||
// Add context if present
|
||||
if (request.context) {
|
||||
allMessages.push({
|
||||
role: 'user',
|
||||
content: request.context,
|
||||
})
|
||||
}
|
||||
|
||||
// Add remaining messages
|
||||
if (request.messages) {
|
||||
allMessages.push(...request.messages)
|
||||
}
|
||||
|
||||
// Transform tools to OpenAI format if provided
|
||||
const tools = request.tools?.length
|
||||
? request.tools.map((tool) => ({
|
||||
type: 'function',
|
||||
function: {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
parameters: tool.parameters,
|
||||
},
|
||||
}))
|
||||
: undefined
|
||||
|
||||
// Build the request payload
|
||||
const payload: any = {
|
||||
model: request.model || 'gemini-2.0-flash',
|
||||
messages: allMessages,
|
||||
}
|
||||
|
||||
// Add optional parameters
|
||||
if (request.temperature !== undefined) payload.temperature = request.temperature
|
||||
if (request.maxTokens !== undefined) payload.max_tokens = request.maxTokens
|
||||
|
||||
// Add response format for structured output if specified
|
||||
if (request.responseFormat) {
|
||||
payload.response_format = {
|
||||
type: 'json_schema',
|
||||
schema: request.responseFormat.schema || request.responseFormat,
|
||||
}
|
||||
}
|
||||
|
||||
// Add tools if provided
|
||||
if (tools?.length) {
|
||||
payload.tools = tools
|
||||
payload.tool_choice = 'auto'
|
||||
}
|
||||
|
||||
// Make the initial API request
|
||||
let currentResponse = await openai.chat.completions.create(payload)
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.prompt_tokens || 0,
|
||||
completion: currentResponse.usage?.completion_tokens || 0,
|
||||
total: currentResponse.usage?.total_tokens || 0,
|
||||
}
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...allMessages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
// Start execution timer for the entire provider execution
|
||||
const providerStartTime = Date.now()
|
||||
const providerStartTimeISO = new Date(providerStartTime).toISOString()
|
||||
|
||||
try {
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
// Check for tool calls
|
||||
const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
|
||||
if (!toolCallsInResponse || toolCallsInResponse.length === 0) {
|
||||
break
|
||||
const openai = new OpenAI({
|
||||
apiKey: request.apiKey,
|
||||
baseURL: 'https://generativelanguage.googleapis.com/v1beta/openai/',
|
||||
dangerouslyAllowBrowser: true,
|
||||
})
|
||||
|
||||
// Start with an empty array for all messages
|
||||
const allMessages = []
|
||||
|
||||
// Add system prompt if present
|
||||
if (request.systemPrompt) {
|
||||
allMessages.push({
|
||||
role: 'system',
|
||||
content: request.systemPrompt,
|
||||
})
|
||||
}
|
||||
|
||||
// Add context if present
|
||||
if (request.context) {
|
||||
allMessages.push({
|
||||
role: 'user',
|
||||
content: request.context,
|
||||
})
|
||||
}
|
||||
|
||||
// Add remaining messages
|
||||
if (request.messages) {
|
||||
allMessages.push(...request.messages)
|
||||
}
|
||||
|
||||
// Transform tools to OpenAI format if provided
|
||||
const tools = request.tools?.length
|
||||
? request.tools.map((tool) => ({
|
||||
type: 'function',
|
||||
function: {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
parameters: tool.parameters,
|
||||
},
|
||||
}))
|
||||
: undefined
|
||||
|
||||
// Build the request payload
|
||||
const payload: any = {
|
||||
model: request.model || 'gemini-2.0-flash',
|
||||
messages: allMessages,
|
||||
}
|
||||
|
||||
// Add optional parameters
|
||||
if (request.temperature !== undefined) payload.temperature = request.temperature
|
||||
if (request.maxTokens !== undefined) payload.max_tokens = request.maxTokens
|
||||
|
||||
// Add response format for structured output if specified
|
||||
if (request.responseFormat) {
|
||||
payload.response_format = {
|
||||
type: 'json_schema',
|
||||
schema: request.responseFormat.schema || request.responseFormat,
|
||||
}
|
||||
}
|
||||
|
||||
// Process each tool call
|
||||
for (const toolCall of toolCallsInResponse) {
|
||||
try {
|
||||
const toolName = toolCall.function.name
|
||||
const toolArgs = JSON.parse(toolCall.function.arguments)
|
||||
// Add tools if provided
|
||||
if (tools?.length) {
|
||||
payload.tools = tools
|
||||
payload.tool_choice = 'auto'
|
||||
}
|
||||
|
||||
// Get the tool from the tools registry
|
||||
const tool = request.tools?.find((t) => t.id === toolName)
|
||||
if (!tool) continue
|
||||
// Make the initial API request
|
||||
const initialCallTime = Date.now()
|
||||
let currentResponse = await openai.chat.completions.create(payload)
|
||||
const firstResponseTime = Date.now() - initialCallTime
|
||||
|
||||
// Execute the tool
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.prompt_tokens || 0,
|
||||
completion: currentResponse.usage?.completion_tokens || 0,
|
||||
total: currentResponse.usage?.total_tokens || 0,
|
||||
}
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...allMessages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
|
||||
if (!result.success) continue
|
||||
// Track time spent in model vs tools
|
||||
let modelTime = firstResponseTime
|
||||
let toolsTime = 0
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: result.timing?.startTime,
|
||||
endTime: result.timing?.endTime,
|
||||
duration: result.timing?.duration,
|
||||
result: result.output,
|
||||
})
|
||||
// Track each model and tool call segment with timestamps
|
||||
const timeSegments: TimeSegment[] = [
|
||||
{
|
||||
type: 'model',
|
||||
name: 'Initial response',
|
||||
startTime: initialCallTime,
|
||||
endTime: initialCallTime + firstResponseTime,
|
||||
duration: firstResponseTime,
|
||||
},
|
||||
]
|
||||
|
||||
// Add the tool call and result to messages
|
||||
currentMessages.push({
|
||||
role: 'assistant',
|
||||
content: null,
|
||||
tool_calls: [
|
||||
{
|
||||
id: toolCall.id,
|
||||
type: 'function',
|
||||
function: {
|
||||
name: toolName,
|
||||
arguments: toolCall.function.arguments,
|
||||
},
|
||||
},
|
||||
],
|
||||
})
|
||||
|
||||
const toolResultContent = JSON.stringify(result.output)
|
||||
|
||||
currentMessages.push({
|
||||
role: 'tool',
|
||||
tool_call_id: toolCall.id,
|
||||
content: toolResultContent,
|
||||
})
|
||||
} catch (error) {
|
||||
logger.error('Error processing tool call:', { error })
|
||||
try {
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
// Check for tool calls
|
||||
const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
|
||||
if (!toolCallsInResponse || toolCallsInResponse.length === 0) {
|
||||
break
|
||||
}
|
||||
|
||||
// Track time for tool calls in this batch
|
||||
const toolsStartTime = Date.now()
|
||||
|
||||
// Process each tool call
|
||||
for (const toolCall of toolCallsInResponse) {
|
||||
try {
|
||||
const toolName = toolCall.function.name
|
||||
const toolArgs = JSON.parse(toolCall.function.arguments)
|
||||
|
||||
// Get the tool from the tools registry
|
||||
const tool = request.tools?.find((t) => t.id === toolName)
|
||||
if (!tool) continue
|
||||
|
||||
// Execute the tool
|
||||
const toolCallStartTime = Date.now()
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
const toolCallEndTime = Date.now()
|
||||
const toolCallDuration = toolCallEndTime - toolCallStartTime
|
||||
|
||||
if (!result.success) continue
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'tool',
|
||||
name: toolName,
|
||||
startTime: toolCallStartTime,
|
||||
endTime: toolCallEndTime,
|
||||
duration: toolCallDuration,
|
||||
})
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: new Date(toolCallStartTime).toISOString(),
|
||||
endTime: new Date(toolCallEndTime).toISOString(),
|
||||
duration: toolCallDuration,
|
||||
result: result.output,
|
||||
})
|
||||
|
||||
// Add the tool call and result to messages
|
||||
currentMessages.push({
|
||||
role: 'assistant',
|
||||
content: null,
|
||||
tool_calls: [
|
||||
{
|
||||
id: toolCall.id,
|
||||
type: 'function',
|
||||
function: {
|
||||
name: toolName,
|
||||
arguments: toolCall.function.arguments,
|
||||
},
|
||||
},
|
||||
],
|
||||
})
|
||||
|
||||
const toolResultContent = JSON.stringify(result.output)
|
||||
|
||||
currentMessages.push({
|
||||
role: 'tool',
|
||||
tool_call_id: toolCall.id,
|
||||
content: toolResultContent,
|
||||
})
|
||||
} catch (error) {
|
||||
logger.error('Error processing tool call:', { error })
|
||||
}
|
||||
}
|
||||
|
||||
// Calculate tool call time for this iteration
|
||||
const thisToolsTime = Date.now() - toolsStartTime
|
||||
toolsTime += thisToolsTime
|
||||
|
||||
// Make the next request with updated messages
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
|
||||
// Time the next model call
|
||||
const nextModelStartTime = Date.now()
|
||||
|
||||
// Make the next request
|
||||
currentResponse = await openai.chat.completions.create(nextPayload)
|
||||
|
||||
const nextModelEndTime = Date.now()
|
||||
const thisModelTime = nextModelEndTime - nextModelStartTime
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'model',
|
||||
name: `Model response (iteration ${iterationCount + 1})`,
|
||||
startTime: nextModelStartTime,
|
||||
endTime: nextModelEndTime,
|
||||
duration: thisModelTime,
|
||||
})
|
||||
|
||||
// Add to model time
|
||||
modelTime += thisModelTime
|
||||
|
||||
// Update content if we have a text response
|
||||
if (currentResponse.choices[0]?.message?.content) {
|
||||
content = currentResponse.choices[0].message.content
|
||||
}
|
||||
|
||||
// Update token counts
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.prompt_tokens || 0
|
||||
tokens.completion += currentResponse.usage.completion_tokens || 0
|
||||
tokens.total += currentResponse.usage.total_tokens || 0
|
||||
}
|
||||
|
||||
iterationCount++
|
||||
}
|
||||
} catch (error) {
|
||||
logger.error('Error in Google Gemini request:', { error })
|
||||
}
|
||||
|
||||
// Make the next request with updated messages
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
// Calculate overall timing
|
||||
const providerEndTime = Date.now()
|
||||
const providerEndTimeISO = new Date(providerEndTime).toISOString()
|
||||
const totalDuration = providerEndTime - providerStartTime
|
||||
|
||||
// Make the next request
|
||||
currentResponse = await openai.chat.completions.create(nextPayload)
|
||||
|
||||
// Update content if we have a text response
|
||||
if (currentResponse.choices[0]?.message?.content) {
|
||||
content = currentResponse.choices[0].message.content
|
||||
}
|
||||
|
||||
// Update token counts
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.prompt_tokens || 0
|
||||
tokens.completion += currentResponse.usage.completion_tokens || 0
|
||||
tokens.total += currentResponse.usage.total_tokens || 0
|
||||
}
|
||||
|
||||
iterationCount++
|
||||
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) {
|
||||
logger.error('Error in Google Gemini request:', { error })
|
||||
throw error
|
||||
}
|
||||
// Include timing information even for errors
|
||||
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,
|
||||
logger.error('Error in Google Gemini request:', { error, duration: totalDuration })
|
||||
|
||||
// Create a new error with timing information
|
||||
const enhancedError = new Error(error instanceof Error ? error.message : String(error))
|
||||
// @ts-ignore - Adding timing property to the error
|
||||
enhancedError.timing = {
|
||||
startTime: providerStartTimeISO,
|
||||
endTime: providerEndTimeISO,
|
||||
duration: totalDuration,
|
||||
}
|
||||
|
||||
throw enhancedError
|
||||
}
|
||||
},
|
||||
}
|
||||
|
||||
+251
-158
@@ -1,7 +1,7 @@
|
||||
import { Groq } from 'groq-sdk'
|
||||
import { createLogger } from '@/lib/logs/console-logger'
|
||||
import { executeTool } from '@/tools'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse } from '../types'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse, TimeSegment } from '../types'
|
||||
|
||||
const logger = createLogger('Groq Provider')
|
||||
|
||||
@@ -18,179 +18,272 @@ export const groqProvider: ProviderConfig = {
|
||||
throw new Error('API key is required for Groq')
|
||||
}
|
||||
|
||||
const groq = new Groq({
|
||||
apiKey: request.apiKey,
|
||||
dangerouslyAllowBrowser: true,
|
||||
})
|
||||
|
||||
// Start with an empty array for all messages
|
||||
const allMessages = []
|
||||
|
||||
// Add system prompt if present
|
||||
if (request.systemPrompt) {
|
||||
allMessages.push({
|
||||
role: 'system',
|
||||
content: request.systemPrompt,
|
||||
})
|
||||
}
|
||||
|
||||
// Add context if present
|
||||
if (request.context) {
|
||||
allMessages.push({
|
||||
role: 'user',
|
||||
content: request.context,
|
||||
})
|
||||
}
|
||||
|
||||
// Add remaining messages
|
||||
if (request.messages) {
|
||||
allMessages.push(...request.messages)
|
||||
}
|
||||
|
||||
// Transform tools to function format if provided
|
||||
const tools = request.tools?.length
|
||||
? request.tools.map((tool) => ({
|
||||
type: 'function',
|
||||
function: {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
parameters: tool.parameters,
|
||||
},
|
||||
}))
|
||||
: undefined
|
||||
|
||||
// Build the request payload
|
||||
const payload: any = {
|
||||
model: (request.model || 'groq/llama-3.3-70b-specdec').replace('groq/', ''),
|
||||
messages: allMessages,
|
||||
}
|
||||
|
||||
// Add optional parameters
|
||||
if (request.temperature !== undefined) payload.temperature = request.temperature
|
||||
if (request.maxTokens !== undefined) payload.max_tokens = request.maxTokens
|
||||
|
||||
// Add response format for structured output if specified
|
||||
if (request.responseFormat) {
|
||||
payload.response_format = {
|
||||
type: 'json_schema',
|
||||
schema: request.responseFormat.schema || request.responseFormat,
|
||||
}
|
||||
}
|
||||
|
||||
// Add tools if provided
|
||||
if (tools?.length) {
|
||||
payload.tools = tools
|
||||
payload.tool_choice = 'auto'
|
||||
}
|
||||
|
||||
// Make the initial API request
|
||||
let currentResponse = await groq.chat.completions.create(payload)
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.prompt_tokens || 0,
|
||||
completion: currentResponse.usage?.completion_tokens || 0,
|
||||
total: currentResponse.usage?.total_tokens || 0,
|
||||
}
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...allMessages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
// Start execution timer for the entire provider execution
|
||||
const providerStartTime = Date.now()
|
||||
const providerStartTimeISO = new Date(providerStartTime).toISOString()
|
||||
|
||||
try {
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
// Check for tool calls
|
||||
const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
|
||||
if (!toolCallsInResponse || toolCallsInResponse.length === 0) {
|
||||
break
|
||||
const groq = new Groq({
|
||||
apiKey: request.apiKey,
|
||||
dangerouslyAllowBrowser: true,
|
||||
})
|
||||
|
||||
// Start with an empty array for all messages
|
||||
const allMessages = []
|
||||
|
||||
// Add system prompt if present
|
||||
if (request.systemPrompt) {
|
||||
allMessages.push({
|
||||
role: 'system',
|
||||
content: request.systemPrompt,
|
||||
})
|
||||
}
|
||||
|
||||
// Add context if present
|
||||
if (request.context) {
|
||||
allMessages.push({
|
||||
role: 'user',
|
||||
content: request.context,
|
||||
})
|
||||
}
|
||||
|
||||
// Add remaining messages
|
||||
if (request.messages) {
|
||||
allMessages.push(...request.messages)
|
||||
}
|
||||
|
||||
// Transform tools to function format if provided
|
||||
const tools = request.tools?.length
|
||||
? request.tools.map((tool) => ({
|
||||
type: 'function',
|
||||
function: {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
parameters: tool.parameters,
|
||||
},
|
||||
}))
|
||||
: undefined
|
||||
|
||||
// Build the request payload
|
||||
const payload: any = {
|
||||
model: (request.model || 'groq/llama-3.3-70b-specdec').replace('groq/', ''),
|
||||
messages: allMessages,
|
||||
}
|
||||
|
||||
// Add optional parameters
|
||||
if (request.temperature !== undefined) payload.temperature = request.temperature
|
||||
if (request.maxTokens !== undefined) payload.max_tokens = request.maxTokens
|
||||
|
||||
// Add response format for structured output if specified
|
||||
if (request.responseFormat) {
|
||||
payload.response_format = {
|
||||
type: 'json_schema',
|
||||
schema: request.responseFormat.schema || request.responseFormat,
|
||||
}
|
||||
}
|
||||
|
||||
// Process each tool call
|
||||
for (const toolCall of toolCallsInResponse) {
|
||||
try {
|
||||
const toolName = toolCall.function.name
|
||||
const toolArgs = JSON.parse(toolCall.function.arguments)
|
||||
// Add tools if provided
|
||||
if (tools?.length) {
|
||||
payload.tools = tools
|
||||
payload.tool_choice = 'auto'
|
||||
}
|
||||
|
||||
// Get the tool from the tools registry
|
||||
const tool = request.tools?.find((t) => t.id === toolName)
|
||||
if (!tool) continue
|
||||
// Make the initial API request
|
||||
const initialCallTime = Date.now()
|
||||
let currentResponse = await groq.chat.completions.create(payload)
|
||||
const firstResponseTime = Date.now() - initialCallTime
|
||||
|
||||
// Execute the tool
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.prompt_tokens || 0,
|
||||
completion: currentResponse.usage?.completion_tokens || 0,
|
||||
total: currentResponse.usage?.total_tokens || 0,
|
||||
}
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...allMessages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
|
||||
if (!result.success) continue
|
||||
// Track time spent in model vs tools
|
||||
let modelTime = firstResponseTime
|
||||
let toolsTime = 0
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: result.timing?.startTime,
|
||||
endTime: result.timing?.endTime,
|
||||
duration: result.timing?.duration,
|
||||
result: result.output,
|
||||
})
|
||||
// Track each model and tool call segment with timestamps
|
||||
const timeSegments: TimeSegment[] = [
|
||||
{
|
||||
type: 'model',
|
||||
name: 'Initial response',
|
||||
startTime: initialCallTime,
|
||||
endTime: initialCallTime + firstResponseTime,
|
||||
duration: firstResponseTime,
|
||||
},
|
||||
]
|
||||
|
||||
// Add the tool call and result to messages
|
||||
currentMessages.push({
|
||||
role: 'assistant',
|
||||
content: null,
|
||||
tool_calls: [
|
||||
{
|
||||
id: toolCall.id,
|
||||
type: 'function',
|
||||
function: {
|
||||
name: toolName,
|
||||
arguments: toolCall.function.arguments,
|
||||
},
|
||||
},
|
||||
],
|
||||
})
|
||||
|
||||
currentMessages.push({
|
||||
role: 'tool',
|
||||
tool_call_id: toolCall.id,
|
||||
content: JSON.stringify(result.output),
|
||||
})
|
||||
} catch (error) {
|
||||
logger.error('Error processing tool call:', { error })
|
||||
try {
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
// Check for tool calls
|
||||
const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
|
||||
if (!toolCallsInResponse || toolCallsInResponse.length === 0) {
|
||||
break
|
||||
}
|
||||
|
||||
// Track time for tool calls in this batch
|
||||
const toolsStartTime = Date.now()
|
||||
|
||||
// Process each tool call
|
||||
for (const toolCall of toolCallsInResponse) {
|
||||
try {
|
||||
const toolName = toolCall.function.name
|
||||
const toolArgs = JSON.parse(toolCall.function.arguments)
|
||||
|
||||
// Get the tool from the tools registry
|
||||
const tool = request.tools?.find((t) => t.id === toolName)
|
||||
if (!tool) continue
|
||||
|
||||
// Execute the tool
|
||||
const toolCallStartTime = Date.now()
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
const toolCallEndTime = Date.now()
|
||||
const toolCallDuration = toolCallEndTime - toolCallStartTime
|
||||
|
||||
if (!result.success) continue
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'tool',
|
||||
name: toolName,
|
||||
startTime: toolCallStartTime,
|
||||
endTime: toolCallEndTime,
|
||||
duration: toolCallDuration,
|
||||
})
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: new Date(toolCallStartTime).toISOString(),
|
||||
endTime: new Date(toolCallEndTime).toISOString(),
|
||||
duration: toolCallDuration,
|
||||
result: result.output,
|
||||
})
|
||||
|
||||
// Add the tool call and result to messages
|
||||
currentMessages.push({
|
||||
role: 'assistant',
|
||||
content: null,
|
||||
tool_calls: [
|
||||
{
|
||||
id: toolCall.id,
|
||||
type: 'function',
|
||||
function: {
|
||||
name: toolName,
|
||||
arguments: toolCall.function.arguments,
|
||||
},
|
||||
},
|
||||
],
|
||||
})
|
||||
|
||||
currentMessages.push({
|
||||
role: 'tool',
|
||||
tool_call_id: toolCall.id,
|
||||
content: JSON.stringify(result.output),
|
||||
})
|
||||
} catch (error) {
|
||||
logger.error('Error processing tool call:', { error })
|
||||
}
|
||||
}
|
||||
|
||||
// Calculate tool call time for this iteration
|
||||
const thisToolsTime = Date.now() - toolsStartTime
|
||||
toolsTime += thisToolsTime
|
||||
|
||||
// Make the next request with updated messages
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
|
||||
// Time the next model call
|
||||
const nextModelStartTime = Date.now()
|
||||
|
||||
// Make the next request
|
||||
currentResponse = await groq.chat.completions.create(nextPayload)
|
||||
|
||||
const nextModelEndTime = Date.now()
|
||||
const thisModelTime = nextModelEndTime - nextModelStartTime
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'model',
|
||||
name: `Model response (iteration ${iterationCount + 1})`,
|
||||
startTime: nextModelStartTime,
|
||||
endTime: nextModelEndTime,
|
||||
duration: thisModelTime,
|
||||
})
|
||||
|
||||
// Add to model time
|
||||
modelTime += thisModelTime
|
||||
|
||||
// Update content if we have a text response
|
||||
if (currentResponse.choices[0]?.message?.content) {
|
||||
content = currentResponse.choices[0].message.content
|
||||
}
|
||||
|
||||
// Update token counts
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.prompt_tokens || 0
|
||||
tokens.completion += currentResponse.usage.completion_tokens || 0
|
||||
tokens.total += currentResponse.usage.total_tokens || 0
|
||||
}
|
||||
|
||||
iterationCount++
|
||||
}
|
||||
} catch (error) {
|
||||
logger.error('Error in Groq request:', { error })
|
||||
}
|
||||
|
||||
// Make the next request with updated messages
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
// Calculate overall timing
|
||||
const providerEndTime = Date.now()
|
||||
const providerEndTimeISO = new Date(providerEndTime).toISOString()
|
||||
const totalDuration = providerEndTime - providerStartTime
|
||||
|
||||
// Make the next request
|
||||
currentResponse = await groq.chat.completions.create(nextPayload)
|
||||
|
||||
// Update content if we have a text response
|
||||
if (currentResponse.choices[0]?.message?.content) {
|
||||
content = currentResponse.choices[0].message.content
|
||||
}
|
||||
|
||||
// Update token counts
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.prompt_tokens || 0
|
||||
tokens.completion += currentResponse.usage.completion_tokens || 0
|
||||
tokens.total += currentResponse.usage.total_tokens || 0
|
||||
}
|
||||
|
||||
iterationCount++
|
||||
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) {
|
||||
logger.error('Error in Groq request:', { error })
|
||||
throw error
|
||||
}
|
||||
// Include timing information even for errors
|
||||
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,
|
||||
logger.error('Error in Groq request:', { error, duration: totalDuration })
|
||||
|
||||
// Create a new error with timing information
|
||||
const enhancedError = new Error(error instanceof Error ? error.message : String(error))
|
||||
// @ts-ignore - Adding timing property to the error
|
||||
enhancedError.timing = {
|
||||
startTime: providerStartTimeISO,
|
||||
endTime: providerEndTimeISO,
|
||||
duration: totalDuration,
|
||||
}
|
||||
|
||||
throw enhancedError
|
||||
}
|
||||
},
|
||||
}
|
||||
|
||||
@@ -1,7 +1,7 @@
|
||||
import OpenAI from 'openai'
|
||||
import { createLogger } from '@/lib/logs/console-logger'
|
||||
import { executeTool } from '@/tools'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse } from '../types'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse, TimeSegment } from '../types'
|
||||
|
||||
const logger = createLogger('OpenAI Provider')
|
||||
|
||||
@@ -106,21 +106,43 @@ export const openaiProvider: ProviderConfig = {
|
||||
logger.info(`Configured ${tools.length} tools for OpenAI request`)
|
||||
}
|
||||
|
||||
// Make the initial API request
|
||||
let currentResponse = await openai.chat.completions.create(payload)
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.prompt_tokens || 0,
|
||||
completion: currentResponse.usage?.completion_tokens || 0,
|
||||
total: currentResponse.usage?.total_tokens || 0,
|
||||
}
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...allMessages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
// Start execution timer for the entire provider execution
|
||||
const providerStartTime = Date.now()
|
||||
const providerStartTimeISO = new Date(providerStartTime).toISOString()
|
||||
|
||||
try {
|
||||
// Make the initial API request
|
||||
const initialCallTime = Date.now()
|
||||
let currentResponse = await openai.chat.completions.create(payload)
|
||||
const firstResponseTime = Date.now() - initialCallTime
|
||||
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.prompt_tokens || 0,
|
||||
completion: currentResponse.usage?.completion_tokens || 0,
|
||||
total: currentResponse.usage?.total_tokens || 0,
|
||||
}
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...allMessages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10 // Prevent infinite loops
|
||||
|
||||
// Track time spent in model vs tools
|
||||
let modelTime = firstResponseTime
|
||||
let toolsTime = 0
|
||||
|
||||
// Track each model and tool call segment with timestamps
|
||||
const timeSegments: TimeSegment[] = [
|
||||
{
|
||||
type: 'model',
|
||||
name: 'Initial response',
|
||||
startTime: initialCallTime,
|
||||
endTime: initialCallTime + firstResponseTime,
|
||||
duration: firstResponseTime,
|
||||
},
|
||||
]
|
||||
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
// Check for tool calls
|
||||
const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
|
||||
@@ -132,6 +154,9 @@ export const openaiProvider: ProviderConfig = {
|
||||
`Processing ${toolCallsInResponse.length} tool calls (iteration ${iterationCount + 1}/${MAX_ITERATIONS})`
|
||||
)
|
||||
|
||||
// Track time for tool calls in this batch
|
||||
const toolsStartTime = Date.now()
|
||||
|
||||
// Process each tool call
|
||||
for (const toolCall of toolCallsInResponse) {
|
||||
try {
|
||||
@@ -143,18 +168,30 @@ export const openaiProvider: ProviderConfig = {
|
||||
if (!tool) continue
|
||||
|
||||
// Execute the tool
|
||||
const toolCallStartTime = Date.now()
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
const toolCallEndTime = Date.now()
|
||||
const toolCallDuration = toolCallEndTime - toolCallStartTime
|
||||
|
||||
if (!result.success) continue
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'tool',
|
||||
name: toolName,
|
||||
startTime: toolCallStartTime,
|
||||
endTime: toolCallEndTime,
|
||||
duration: toolCallDuration,
|
||||
})
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: result.timing?.startTime,
|
||||
endTime: result.timing?.endTime,
|
||||
duration: result.timing?.duration,
|
||||
startTime: new Date(toolCallStartTime).toISOString(),
|
||||
endTime: new Date(toolCallEndTime).toISOString(),
|
||||
duration: toolCallDuration,
|
||||
result: result.output,
|
||||
})
|
||||
|
||||
@@ -187,15 +224,37 @@ export const openaiProvider: ProviderConfig = {
|
||||
}
|
||||
}
|
||||
|
||||
// Calculate tool call time for this iteration
|
||||
const thisToolsTime = Date.now() - toolsStartTime
|
||||
toolsTime += thisToolsTime
|
||||
|
||||
// Make the next request with updated messages
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
|
||||
// Time the next model call
|
||||
const nextModelStartTime = Date.now()
|
||||
|
||||
// Make the next request
|
||||
currentResponse = await openai.chat.completions.create(nextPayload)
|
||||
|
||||
const nextModelEndTime = Date.now()
|
||||
const thisModelTime = nextModelEndTime - nextModelStartTime
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'model',
|
||||
name: `Model response (iteration ${iterationCount + 1})`,
|
||||
startTime: nextModelStartTime,
|
||||
endTime: nextModelEndTime,
|
||||
duration: thisModelTime,
|
||||
})
|
||||
|
||||
// Add to model time
|
||||
modelTime += thisModelTime
|
||||
|
||||
// Update content if we have a text response
|
||||
if (currentResponse.choices[0]?.message?.content) {
|
||||
content = currentResponse.choices[0].message.content
|
||||
@@ -210,17 +269,47 @@ export const openaiProvider: ProviderConfig = {
|
||||
|
||||
iterationCount++
|
||||
}
|
||||
} catch (error) {
|
||||
logger.error('Error in OpenAI request:', { error })
|
||||
throw error
|
||||
}
|
||||
|
||||
return {
|
||||
content,
|
||||
model: request.model,
|
||||
tokens,
|
||||
toolCalls: toolCalls.length > 0 ? toolCalls : undefined,
|
||||
toolResults: toolResults.length > 0 ? toolResults : undefined,
|
||||
// Calculate overall timing
|
||||
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) {
|
||||
// Include timing information even for errors
|
||||
const providerEndTime = Date.now()
|
||||
const providerEndTimeISO = new Date(providerEndTime).toISOString()
|
||||
const totalDuration = providerEndTime - providerStartTime
|
||||
|
||||
logger.error('Error in OpenAI request:', { error, duration: totalDuration })
|
||||
|
||||
// Create a new error with timing information
|
||||
const enhancedError = new Error(error instanceof Error ? error.message : String(error))
|
||||
// @ts-ignore - Adding timing property to the error
|
||||
enhancedError.timing = {
|
||||
startTime: providerStartTimeISO,
|
||||
endTime: providerEndTimeISO,
|
||||
duration: totalDuration,
|
||||
}
|
||||
|
||||
throw enhancedError
|
||||
}
|
||||
},
|
||||
}
|
||||
|
||||
@@ -39,6 +39,14 @@ export interface FunctionCallResponse {
|
||||
input?: Record<string, any>
|
||||
}
|
||||
|
||||
export interface TimeSegment {
|
||||
type: 'model' | 'tool'
|
||||
name: string
|
||||
startTime: number
|
||||
endTime: number
|
||||
duration: number
|
||||
}
|
||||
|
||||
export interface ProviderResponse {
|
||||
content: string
|
||||
model: string
|
||||
@@ -49,6 +57,16 @@ export interface ProviderResponse {
|
||||
}
|
||||
toolCalls?: FunctionCallResponse[]
|
||||
toolResults?: any[]
|
||||
timing?: {
|
||||
startTime: string // ISO timestamp when provider execution started
|
||||
endTime: string // ISO timestamp when provider execution completed
|
||||
duration: number // Total duration in milliseconds
|
||||
modelTime?: number // Time spent in model generation (excluding tool calls)
|
||||
toolsTime?: number // Time spent in tool calls
|
||||
firstResponseTime?: number // Time to first token/response
|
||||
iterations?: number // Number of model calls for tool use
|
||||
timeSegments?: TimeSegment[] // Detailed timeline of all operations
|
||||
}
|
||||
}
|
||||
|
||||
export interface ProviderToolConfig {
|
||||
|
||||
+246
-152
@@ -1,7 +1,7 @@
|
||||
import OpenAI from 'openai'
|
||||
import { createLogger } from '@/lib/logs/console-logger'
|
||||
import { executeTool } from '@/tools'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse } from '../types'
|
||||
import { ProviderConfig, ProviderRequest, ProviderResponse, TimeSegment } from '../types'
|
||||
|
||||
const logger = createLogger('XAI Provider')
|
||||
|
||||
@@ -18,174 +18,268 @@ export const xAIProvider: ProviderConfig = {
|
||||
throw new Error('API key is required for xAI')
|
||||
}
|
||||
|
||||
const xai = new OpenAI({
|
||||
apiKey: request.apiKey,
|
||||
baseURL: 'https://api.x.ai/v1',
|
||||
dangerouslyAllowBrowser: true,
|
||||
})
|
||||
|
||||
const allMessages = []
|
||||
|
||||
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 tools = request.tools?.length
|
||||
? request.tools.map((tool) => ({
|
||||
type: 'function',
|
||||
function: {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
parameters: tool.parameters,
|
||||
},
|
||||
}))
|
||||
: undefined
|
||||
|
||||
const payload: any = {
|
||||
model: request.model || 'grok-2-latest',
|
||||
messages: allMessages,
|
||||
}
|
||||
|
||||
if (request.temperature !== undefined) payload.temperature = request.temperature
|
||||
if (request.maxTokens !== undefined) payload.max_tokens = request.maxTokens
|
||||
|
||||
if (request.responseFormat) {
|
||||
payload.response_format = {
|
||||
type: 'json_schema',
|
||||
json_schema: {
|
||||
name: request.responseFormat.name || 'structured_response',
|
||||
schema: request.responseFormat.schema || request.responseFormat,
|
||||
strict: request.responseFormat.strict !== false,
|
||||
},
|
||||
}
|
||||
|
||||
if (allMessages.length > 0 && allMessages[0].role === 'system') {
|
||||
allMessages[0].content = `${allMessages[0].content}\n\nYou MUST respond with a valid JSON object. DO NOT include any other text, explanations, or markdown formatting in your response - ONLY the JSON object.`
|
||||
} else {
|
||||
allMessages.unshift({
|
||||
role: 'system',
|
||||
content: `You MUST respond with a valid JSON object. DO NOT include any other text, explanations, or markdown formatting in your response - ONLY the JSON object.`,
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
if (tools?.length) {
|
||||
payload.tools = tools
|
||||
payload.tool_choice = 'auto'
|
||||
}
|
||||
|
||||
let currentResponse = await xai.chat.completions.create(payload)
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.prompt_tokens || 0,
|
||||
completion: currentResponse.usage?.completion_tokens || 0,
|
||||
total: currentResponse.usage?.total_tokens || 0,
|
||||
}
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...allMessages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10
|
||||
// Start execution timer for the entire provider execution
|
||||
const providerStartTime = Date.now()
|
||||
const providerStartTimeISO = new Date(providerStartTime).toISOString()
|
||||
|
||||
try {
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
|
||||
if (!toolCallsInResponse || toolCallsInResponse.length === 0) {
|
||||
break
|
||||
const xai = new OpenAI({
|
||||
apiKey: request.apiKey,
|
||||
baseURL: 'https://api.x.ai/v1',
|
||||
dangerouslyAllowBrowser: true,
|
||||
})
|
||||
|
||||
const allMessages = []
|
||||
|
||||
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 tools = request.tools?.length
|
||||
? request.tools.map((tool) => ({
|
||||
type: 'function',
|
||||
function: {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
parameters: tool.parameters,
|
||||
},
|
||||
}))
|
||||
: undefined
|
||||
|
||||
const payload: any = {
|
||||
model: request.model || 'grok-2-latest',
|
||||
messages: allMessages,
|
||||
}
|
||||
|
||||
if (request.temperature !== undefined) payload.temperature = request.temperature
|
||||
if (request.maxTokens !== undefined) payload.max_tokens = request.maxTokens
|
||||
|
||||
if (request.responseFormat) {
|
||||
payload.response_format = {
|
||||
type: 'json_schema',
|
||||
json_schema: {
|
||||
name: request.responseFormat.name || 'structured_response',
|
||||
schema: request.responseFormat.schema || request.responseFormat,
|
||||
strict: request.responseFormat.strict !== false,
|
||||
},
|
||||
}
|
||||
|
||||
for (const toolCall of toolCallsInResponse) {
|
||||
try {
|
||||
const toolName = toolCall.function.name
|
||||
const toolArgs = JSON.parse(toolCall.function.arguments)
|
||||
if (allMessages.length > 0 && allMessages[0].role === 'system') {
|
||||
allMessages[0].content = `${allMessages[0].content}\n\nYou MUST respond with a valid JSON object. DO NOT include any other text, explanations, or markdown formatting in your response - ONLY the JSON object.`
|
||||
} else {
|
||||
allMessages.unshift({
|
||||
role: 'system',
|
||||
content: `You MUST respond with a valid JSON object. DO NOT include any other text, explanations, or markdown formatting in your response - ONLY the JSON object.`,
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
const tool = request.tools?.find((t) => t.id === toolName)
|
||||
if (!tool) continue
|
||||
if (tools?.length) {
|
||||
payload.tools = tools
|
||||
payload.tool_choice = 'auto'
|
||||
}
|
||||
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
// Make the initial API request
|
||||
const initialCallTime = Date.now()
|
||||
let currentResponse = await xai.chat.completions.create(payload)
|
||||
const firstResponseTime = Date.now() - initialCallTime
|
||||
|
||||
if (!result.success) continue
|
||||
let content = currentResponse.choices[0]?.message?.content || ''
|
||||
let tokens = {
|
||||
prompt: currentResponse.usage?.prompt_tokens || 0,
|
||||
completion: currentResponse.usage?.completion_tokens || 0,
|
||||
total: currentResponse.usage?.total_tokens || 0,
|
||||
}
|
||||
let toolCalls = []
|
||||
let toolResults = []
|
||||
let currentMessages = [...allMessages]
|
||||
let iterationCount = 0
|
||||
const MAX_ITERATIONS = 10
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: result.timing?.startTime,
|
||||
endTime: result.timing?.endTime,
|
||||
duration: result.timing?.duration,
|
||||
result: result.output,
|
||||
})
|
||||
// Track time spent in model vs tools
|
||||
let modelTime = firstResponseTime
|
||||
let toolsTime = 0
|
||||
|
||||
currentMessages.push({
|
||||
role: 'assistant',
|
||||
content: null,
|
||||
tool_calls: [
|
||||
{
|
||||
id: toolCall.id,
|
||||
type: 'function',
|
||||
function: {
|
||||
name: toolName,
|
||||
arguments: toolCall.function.arguments,
|
||||
},
|
||||
},
|
||||
],
|
||||
})
|
||||
// Track each model and tool call segment with timestamps
|
||||
const timeSegments: TimeSegment[] = [
|
||||
{
|
||||
type: 'model',
|
||||
name: 'Initial response',
|
||||
startTime: initialCallTime,
|
||||
endTime: initialCallTime + firstResponseTime,
|
||||
duration: firstResponseTime,
|
||||
},
|
||||
]
|
||||
|
||||
currentMessages.push({
|
||||
role: 'tool',
|
||||
tool_call_id: toolCall.id,
|
||||
content: JSON.stringify(result.output),
|
||||
})
|
||||
} catch (error) {
|
||||
logger.error('Error processing tool call:', { error })
|
||||
try {
|
||||
while (iterationCount < MAX_ITERATIONS) {
|
||||
const toolCallsInResponse = currentResponse.choices[0]?.message?.tool_calls
|
||||
if (!toolCallsInResponse || toolCallsInResponse.length === 0) {
|
||||
break
|
||||
}
|
||||
|
||||
// Track time for tool calls in this batch
|
||||
const toolsStartTime = Date.now()
|
||||
|
||||
for (const toolCall of toolCallsInResponse) {
|
||||
try {
|
||||
const toolName = toolCall.function.name
|
||||
const toolArgs = JSON.parse(toolCall.function.arguments)
|
||||
|
||||
const tool = request.tools?.find((t) => t.id === toolName)
|
||||
if (!tool) continue
|
||||
|
||||
const toolCallStartTime = Date.now()
|
||||
const mergedArgs = { ...tool.params, ...toolArgs }
|
||||
const result = await executeTool(toolName, mergedArgs)
|
||||
const toolCallEndTime = Date.now()
|
||||
const toolCallDuration = toolCallEndTime - toolCallStartTime
|
||||
|
||||
if (!result.success) continue
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'tool',
|
||||
name: toolName,
|
||||
startTime: toolCallStartTime,
|
||||
endTime: toolCallEndTime,
|
||||
duration: toolCallDuration,
|
||||
})
|
||||
|
||||
toolResults.push(result.output)
|
||||
toolCalls.push({
|
||||
name: toolName,
|
||||
arguments: toolArgs,
|
||||
startTime: new Date(toolCallStartTime).toISOString(),
|
||||
endTime: new Date(toolCallEndTime).toISOString(),
|
||||
duration: toolCallDuration,
|
||||
result: result.output,
|
||||
})
|
||||
|
||||
currentMessages.push({
|
||||
role: 'assistant',
|
||||
content: null,
|
||||
tool_calls: [
|
||||
{
|
||||
id: toolCall.id,
|
||||
type: 'function',
|
||||
function: {
|
||||
name: toolName,
|
||||
arguments: toolCall.function.arguments,
|
||||
},
|
||||
},
|
||||
],
|
||||
})
|
||||
|
||||
currentMessages.push({
|
||||
role: 'tool',
|
||||
tool_call_id: toolCall.id,
|
||||
content: JSON.stringify(result.output),
|
||||
})
|
||||
} catch (error) {
|
||||
logger.error('Error processing tool call:', { error })
|
||||
}
|
||||
}
|
||||
|
||||
// Calculate tool call time for this iteration
|
||||
const thisToolsTime = Date.now() - toolsStartTime
|
||||
toolsTime += thisToolsTime
|
||||
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
|
||||
// Time the next model call
|
||||
const nextModelStartTime = Date.now()
|
||||
|
||||
currentResponse = await xai.chat.completions.create(nextPayload)
|
||||
|
||||
const nextModelEndTime = Date.now()
|
||||
const thisModelTime = nextModelEndTime - nextModelStartTime
|
||||
|
||||
// Add to time segments
|
||||
timeSegments.push({
|
||||
type: 'model',
|
||||
name: `Model response (iteration ${iterationCount + 1})`,
|
||||
startTime: nextModelStartTime,
|
||||
endTime: nextModelEndTime,
|
||||
duration: thisModelTime,
|
||||
})
|
||||
|
||||
// Add to model time
|
||||
modelTime += thisModelTime
|
||||
|
||||
if (currentResponse.choices[0]?.message?.content) {
|
||||
content = currentResponse.choices[0].message.content
|
||||
}
|
||||
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.prompt_tokens || 0
|
||||
tokens.completion += currentResponse.usage.completion_tokens || 0
|
||||
tokens.total += currentResponse.usage.total_tokens || 0
|
||||
}
|
||||
|
||||
iterationCount++
|
||||
}
|
||||
} catch (error) {
|
||||
logger.error('Error in xAI request:', { error })
|
||||
}
|
||||
|
||||
const nextPayload = {
|
||||
...payload,
|
||||
messages: currentMessages,
|
||||
}
|
||||
// Calculate overall timing
|
||||
const providerEndTime = Date.now()
|
||||
const providerEndTimeISO = new Date(providerEndTime).toISOString()
|
||||
const totalDuration = providerEndTime - providerStartTime
|
||||
|
||||
currentResponse = await xai.chat.completions.create(nextPayload)
|
||||
|
||||
if (currentResponse.choices[0]?.message?.content) {
|
||||
content = currentResponse.choices[0].message.content
|
||||
}
|
||||
|
||||
if (currentResponse.usage) {
|
||||
tokens.prompt += currentResponse.usage.prompt_tokens || 0
|
||||
tokens.completion += currentResponse.usage.completion_tokens || 0
|
||||
tokens.total += currentResponse.usage.total_tokens || 0
|
||||
}
|
||||
|
||||
iterationCount++
|
||||
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) {
|
||||
logger.error('Error in xAI request:', { error })
|
||||
throw error
|
||||
}
|
||||
// Include timing information even for errors
|
||||
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,
|
||||
logger.error('Error in xAI request:', { error, duration: totalDuration })
|
||||
|
||||
// Create a new error with timing information
|
||||
const enhancedError = new Error(error instanceof Error ? error.message : String(error))
|
||||
// @ts-ignore - Adding timing property to the error
|
||||
enhancedError.timing = {
|
||||
startTime: providerStartTimeISO,
|
||||
endTime: providerEndTimeISO,
|
||||
duration: totalDuration,
|
||||
}
|
||||
|
||||
throw enhancedError
|
||||
}
|
||||
},
|
||||
}
|
||||
|
||||
@@ -6,15 +6,252 @@ import { useExecutionStore } from '@/stores/execution/store'
|
||||
import { useNotificationStore } from '@/stores/notifications/store'
|
||||
import { useEnvironmentStore } from '@/stores/settings/environment/store'
|
||||
import { useWorkflowRegistry } from '@/stores/workflows/registry/store'
|
||||
import { useSubBlockStore } from '@/stores/workflows/subblock/store'
|
||||
import { mergeSubblockState } from '@/stores/workflows/utils'
|
||||
import { useWorkflowStore } from '@/stores/workflows/workflow/store'
|
||||
import { TraceSpan } from '@/app/w/logs/stores/types'
|
||||
import { Executor } from '@/executor'
|
||||
import { ExecutionResult } from '@/executor/types'
|
||||
import { Serializer } from '@/serializer'
|
||||
|
||||
const logger = createLogger('useWorkflowExecution')
|
||||
|
||||
// Helper function to build a tree of trace spans from execution logs
|
||||
function buildTraceSpans(result: ExecutionResult): {
|
||||
traceSpans: TraceSpan[]
|
||||
totalDuration: number
|
||||
} {
|
||||
// If no logs, return empty spans
|
||||
if (!result.logs || result.logs.length === 0) {
|
||||
return { traceSpans: [], totalDuration: 0 }
|
||||
}
|
||||
|
||||
// Store all spans as a map for faster lookup
|
||||
const spanMap = new Map<string, TraceSpan>()
|
||||
|
||||
// First pass: Create spans for each block
|
||||
result.logs.forEach((log) => {
|
||||
// Skip logs that don't have block execution information
|
||||
if (!log.blockId || !log.blockType) return
|
||||
|
||||
// Create a unique ID for this span using blockId and timestamp
|
||||
const spanId = `${log.blockId}-${new Date(log.startedAt).getTime()}`
|
||||
|
||||
// Extract duration if available
|
||||
const duration = log.durationMs || 0
|
||||
|
||||
// Create the span
|
||||
const span: TraceSpan = {
|
||||
id: spanId,
|
||||
name: log.blockName || log.blockId,
|
||||
type: log.blockType,
|
||||
duration: duration,
|
||||
startTime: log.startedAt,
|
||||
endTime: log.endedAt,
|
||||
status: log.error ? 'error' : 'success',
|
||||
children: [],
|
||||
}
|
||||
|
||||
// Add provider timing data if it exists
|
||||
if (log.output?.response?.providerTiming) {
|
||||
const providerTiming = log.output.response.providerTiming
|
||||
|
||||
// If we have time segments, use them to create a more detailed timeline
|
||||
if (providerTiming.timeSegments && providerTiming.timeSegments.length > 0) {
|
||||
const segmentStartTime = new Date(log.startedAt).getTime()
|
||||
const children: TraceSpan[] = []
|
||||
|
||||
// Process segments in order
|
||||
providerTiming.timeSegments.forEach(
|
||||
(
|
||||
segment: {
|
||||
type: string
|
||||
name: string
|
||||
startTime: number
|
||||
endTime: number
|
||||
duration: number
|
||||
},
|
||||
index: number
|
||||
) => {
|
||||
const relativeStart = segment.startTime - segmentStartTime
|
||||
|
||||
// Enhance the segment name to include model information for model segments
|
||||
let enhancedName = segment.name
|
||||
if (segment.type === 'model') {
|
||||
const modelName = log.output.response.model || ''
|
||||
|
||||
if (segment.name === 'Initial response') {
|
||||
enhancedName = `Initial response${modelName ? ` (${modelName})` : ''}`
|
||||
} else if (segment.name.includes('iteration')) {
|
||||
// Extract the iteration number
|
||||
const iterationMatch = segment.name.match(/\(iteration (\d+)\)/)
|
||||
const iterationNum = iterationMatch ? iterationMatch[1] : ''
|
||||
|
||||
enhancedName = `Model response${iterationNum ? ` (iteration ${iterationNum})` : ''}${modelName ? ` (${modelName})` : ''}`
|
||||
}
|
||||
}
|
||||
|
||||
const segmentSpan: TraceSpan = {
|
||||
id: `${spanId}-segment-${index}`,
|
||||
name: enhancedName,
|
||||
// Make sure we handle model and tool types, and fallback to generic 'span' for anything else
|
||||
type: segment.type === 'model' || segment.type === 'tool' ? segment.type : 'span',
|
||||
duration: segment.duration,
|
||||
startTime: new Date(segment.startTime).toISOString(),
|
||||
endTime: new Date(segment.endTime).toISOString(),
|
||||
status: 'success',
|
||||
// Add relative timing display for segments after the first one
|
||||
relativeStartMs: index === 0 ? undefined : relativeStart,
|
||||
// For model segments, add token info if available
|
||||
...(segment.type === 'model' && {
|
||||
tokens: index === 0 ? log.output.response.tokens?.completion : undefined,
|
||||
}),
|
||||
}
|
||||
|
||||
children.push(segmentSpan)
|
||||
}
|
||||
)
|
||||
|
||||
// Add all segments as children
|
||||
if (!span.children) span.children = []
|
||||
span.children.push(...children)
|
||||
}
|
||||
// If no segments but we have provider timing, create a provider span
|
||||
else {
|
||||
// Create a child span for the provider execution
|
||||
const providerSpan: TraceSpan = {
|
||||
id: `${spanId}-provider`,
|
||||
name: log.output.response.model || 'AI Provider',
|
||||
type: 'provider',
|
||||
duration: providerTiming.duration || 0,
|
||||
startTime: providerTiming.startTime || log.startedAt,
|
||||
endTime: providerTiming.endTime || log.endedAt,
|
||||
status: 'success',
|
||||
tokens: log.output.response.tokens?.total,
|
||||
}
|
||||
|
||||
// If we have model time, create a child span for just the model processing
|
||||
if (providerTiming.modelTime) {
|
||||
const modelName = log.output.response.model || ''
|
||||
const modelSpan: TraceSpan = {
|
||||
id: `${spanId}-model`,
|
||||
name: `Model Generation${modelName ? ` (${modelName})` : ''}`,
|
||||
type: 'model',
|
||||
duration: providerTiming.modelTime,
|
||||
startTime: providerTiming.startTime, // Approximate
|
||||
endTime: providerTiming.endTime, // Approximate
|
||||
status: 'success',
|
||||
tokens: log.output.response.tokens?.completion,
|
||||
}
|
||||
|
||||
if (!providerSpan.children) providerSpan.children = []
|
||||
providerSpan.children.push(modelSpan)
|
||||
}
|
||||
|
||||
if (!span.children) span.children = []
|
||||
span.children.push(providerSpan)
|
||||
|
||||
// When using provider timing without segments, still add tool calls if they exist
|
||||
if (log.output?.response?.toolCalls?.list) {
|
||||
span.toolCalls = log.output.response.toolCalls.list.map((tc: any) => ({
|
||||
name: tc.name,
|
||||
duration: tc.duration || 0,
|
||||
startTime: tc.startTime || log.startedAt,
|
||||
endTime: tc.endTime || log.endedAt,
|
||||
status: tc.error ? 'error' : 'success',
|
||||
input: tc.arguments || tc.input,
|
||||
output: tc.result || tc.output,
|
||||
error: tc.error,
|
||||
}))
|
||||
}
|
||||
}
|
||||
} else {
|
||||
// When not using provider timing at all, add tool calls if they exist
|
||||
if (log.output?.response?.toolCalls?.list) {
|
||||
span.toolCalls = log.output.response.toolCalls.list.map((tc: any) => ({
|
||||
name: tc.name,
|
||||
duration: tc.duration || 0,
|
||||
startTime: tc.startTime || log.startedAt,
|
||||
endTime: tc.endTime || log.endedAt,
|
||||
status: tc.error ? 'error' : 'success',
|
||||
input: tc.arguments || tc.input,
|
||||
output: tc.result || tc.output,
|
||||
error: tc.error,
|
||||
}))
|
||||
}
|
||||
}
|
||||
|
||||
// Store in map
|
||||
spanMap.set(spanId, span)
|
||||
})
|
||||
|
||||
// Second pass: Build the hierarchy
|
||||
// We'll first need to sort logs chronologically
|
||||
const sortedLogs = [...result.logs].sort((a, b) => {
|
||||
const aTime = new Date(a.startedAt).getTime()
|
||||
const bTime = new Date(b.startedAt).getTime()
|
||||
return aTime - bTime
|
||||
})
|
||||
|
||||
// Track parent spans using a stack
|
||||
const spanStack: TraceSpan[] = []
|
||||
const rootSpans: TraceSpan[] = []
|
||||
|
||||
// Process logs to build the hierarchy
|
||||
sortedLogs.forEach((log) => {
|
||||
if (!log.blockId || !log.blockType) return
|
||||
|
||||
const spanId = `${log.blockId}-${new Date(log.startedAt).getTime()}`
|
||||
const span = spanMap.get(spanId)
|
||||
if (!span) return
|
||||
|
||||
// If we have a non-empty stack, check if this span should be a child
|
||||
if (spanStack.length > 0) {
|
||||
const potentialParent = spanStack[spanStack.length - 1]
|
||||
const parentStartTime = new Date(potentialParent.startTime).getTime()
|
||||
const parentEndTime = new Date(potentialParent.endTime).getTime()
|
||||
const spanStartTime = new Date(span.startTime).getTime()
|
||||
|
||||
// If this span starts after the parent starts and the parent is still on the stack,
|
||||
// we'll assume it's a child span
|
||||
if (spanStartTime >= parentStartTime && spanStartTime <= parentEndTime) {
|
||||
if (!potentialParent.children) potentialParent.children = []
|
||||
potentialParent.children.push(span)
|
||||
} else {
|
||||
// This span doesn't belong to the current parent, pop from stack
|
||||
while (
|
||||
spanStack.length > 0 &&
|
||||
new Date(spanStack[spanStack.length - 1].endTime).getTime() < spanStartTime
|
||||
) {
|
||||
spanStack.pop()
|
||||
}
|
||||
|
||||
// Check if we still have a parent
|
||||
if (spanStack.length > 0) {
|
||||
const newParent = spanStack[spanStack.length - 1]
|
||||
if (!newParent.children) newParent.children = []
|
||||
newParent.children.push(span)
|
||||
} else {
|
||||
// No parent, this is a root span
|
||||
rootSpans.push(span)
|
||||
}
|
||||
}
|
||||
} else {
|
||||
// Empty stack, this is a root span
|
||||
rootSpans.push(span)
|
||||
}
|
||||
|
||||
// Check if this span could be a parent to future spans
|
||||
if (log.blockType === 'agent' || log.blockType === 'workflow') {
|
||||
spanStack.push(span)
|
||||
}
|
||||
})
|
||||
|
||||
// Calculate total duration as the sum of root spans
|
||||
const totalDuration = rootSpans.reduce((sum, span) => sum + span.duration, 0)
|
||||
|
||||
return { traceSpans: rootSpans, totalDuration }
|
||||
}
|
||||
|
||||
export function useWorkflowExecution() {
|
||||
const { blocks, edges, loops } = useWorkflowStore()
|
||||
const { activeWorkflowId } = useWorkflowRegistry()
|
||||
@@ -26,6 +263,16 @@ export function useWorkflowExecution() {
|
||||
|
||||
const persistLogs = async (executionId: string, result: ExecutionResult) => {
|
||||
try {
|
||||
// Build trace spans from execution logs
|
||||
const { traceSpans, totalDuration } = buildTraceSpans(result)
|
||||
|
||||
// Add trace spans to the execution result
|
||||
const enrichedResult = {
|
||||
...result,
|
||||
traceSpans,
|
||||
totalDuration,
|
||||
}
|
||||
|
||||
const response = await fetch(`/api/workflow/${activeWorkflowId}/log`, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
@@ -33,7 +280,7 @@ export function useWorkflowExecution() {
|
||||
},
|
||||
body: JSON.stringify({
|
||||
executionId,
|
||||
result,
|
||||
result: enrichedResult,
|
||||
}),
|
||||
})
|
||||
|
||||
|
||||
@@ -8,6 +8,7 @@ import { ScrollArea } from '@/components/ui/scroll-area'
|
||||
import { WorkflowLog } from '@/app/w/logs/stores/types'
|
||||
import { formatDate } from '@/app/w/logs/utils/format-date'
|
||||
import { ToolCallsDisplay } from '../tool-calls/tool-calls-display'
|
||||
import { TraceSpansDisplay } from '../trace-spans/trace-spans-display'
|
||||
|
||||
interface LogSidebarProps {
|
||||
log: WorkflowLog | null
|
||||
@@ -121,6 +122,22 @@ export function Sidebar({ log, isOpen, onClose }: LogSidebarProps) {
|
||||
return formatJsonContent(log.message)
|
||||
}, [log])
|
||||
|
||||
// Determine if this is a workflow execution log
|
||||
const isWorkflowExecutionLog = useMemo(() => {
|
||||
if (!log) return false
|
||||
// Check if message contains "workflow executed" or similar phrases
|
||||
return (
|
||||
log.message.toLowerCase().includes('workflow executed') ||
|
||||
log.message.toLowerCase().includes('execution completed') ||
|
||||
(log.trigger === 'manual' && log.duration)
|
||||
)
|
||||
}, [log])
|
||||
|
||||
// Helper to determine if we have trace spans to display
|
||||
const hasTraceSpans = useMemo(() => {
|
||||
return !!(log?.metadata?.traceSpans && log.metadata.traceSpans.length > 0)
|
||||
}, [log])
|
||||
|
||||
const handleMouseDown = (e: React.MouseEvent) => {
|
||||
setIsDragging(true)
|
||||
e.preventDefault()
|
||||
@@ -262,6 +279,17 @@ export function Sidebar({ log, isOpen, onClose }: LogSidebarProps) {
|
||||
</div>
|
||||
)}
|
||||
|
||||
{/* Trace Spans (if available and this is a workflow execution log) */}
|
||||
{isWorkflowExecutionLog && log.metadata?.traceSpans && (
|
||||
<div>
|
||||
<h3 className="text-xs font-medium text-muted-foreground mb-1">Trace Spans</h3>
|
||||
<TraceSpansDisplay
|
||||
traceSpans={log.metadata.traceSpans}
|
||||
totalDuration={log.metadata.totalDuration}
|
||||
/>
|
||||
</div>
|
||||
)}
|
||||
|
||||
{/* Tool Calls (if available) */}
|
||||
{log.metadata?.toolCalls && log.metadata.toolCalls.length > 0 && (
|
||||
<div>
|
||||
|
||||
@@ -0,0 +1,359 @@
|
||||
'use client'
|
||||
|
||||
import { useState } from 'react'
|
||||
import {
|
||||
BarChart,
|
||||
ChevronDown,
|
||||
ChevronRight,
|
||||
Code,
|
||||
Cpu,
|
||||
Database,
|
||||
ExternalLink,
|
||||
} from 'lucide-react'
|
||||
import {
|
||||
AgentIcon,
|
||||
ApiIcon,
|
||||
ChartBarIcon,
|
||||
CodeIcon,
|
||||
ConditionalIcon,
|
||||
ConnectIcon,
|
||||
} from '@/components/icons'
|
||||
import { cn } from '@/lib/utils'
|
||||
import { TraceSpan } from '../../stores/types'
|
||||
|
||||
interface TraceSpansDisplayProps {
|
||||
traceSpans?: TraceSpan[]
|
||||
totalDuration?: number
|
||||
}
|
||||
|
||||
export function TraceSpansDisplay({ traceSpans, totalDuration = 0 }: TraceSpansDisplayProps) {
|
||||
if (!traceSpans || traceSpans.length === 0) {
|
||||
return <div className="text-sm text-muted-foreground">No trace data available</div>
|
||||
}
|
||||
|
||||
// Format total duration for better readability
|
||||
const formatTotalDuration = (ms: number) => {
|
||||
if (ms < 1000) return `${ms}ms`
|
||||
return `${(ms / 1000).toFixed(2)}s (${ms}ms)`
|
||||
}
|
||||
|
||||
// Find the earliest start time among all spans to be the workflow start time
|
||||
const workflowStartTime = traceSpans.reduce((earliest, span) => {
|
||||
const startTime = new Date(span.startTime).getTime()
|
||||
return startTime < earliest ? startTime : earliest
|
||||
}, Infinity)
|
||||
|
||||
return (
|
||||
<div>
|
||||
{/* Show only the total duration in the header */}
|
||||
<div className="text-xs font-semibold text-right px-2 text-muted-foreground">
|
||||
<span className="font-mono">{formatTotalDuration(totalDuration)}</span>
|
||||
</div>
|
||||
|
||||
<div className="border rounded-md bg-background/40 mt-0.5">
|
||||
{traceSpans.map((span, index) => (
|
||||
<TraceSpanItem
|
||||
key={index}
|
||||
span={span}
|
||||
depth={0}
|
||||
totalDuration={totalDuration}
|
||||
isLast={index === traceSpans.length - 1}
|
||||
parentStartTime={new Date(span.startTime).getTime()}
|
||||
workflowStartTime={workflowStartTime}
|
||||
/>
|
||||
))}
|
||||
</div>
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
||||
interface TraceSpanItemProps {
|
||||
span: TraceSpan
|
||||
depth: number
|
||||
totalDuration: number
|
||||
isLast: boolean
|
||||
parentStartTime: number // Start time of the parent span for offset calculation
|
||||
workflowStartTime: number // Start time of the entire workflow
|
||||
}
|
||||
|
||||
function TraceSpanItem({
|
||||
span,
|
||||
depth,
|
||||
totalDuration,
|
||||
isLast,
|
||||
parentStartTime,
|
||||
workflowStartTime,
|
||||
}: TraceSpanItemProps): JSX.Element {
|
||||
const [expanded, setExpanded] = useState(false) // Collapsed by default
|
||||
const hasChildren = span.children && span.children.length > 0
|
||||
const hasToolCalls = span.toolCalls && span.toolCalls.length > 0
|
||||
|
||||
// Calculate timing information
|
||||
const spanStartTime = new Date(span.startTime).getTime()
|
||||
const spanEndTime = new Date(span.endTime).getTime()
|
||||
const duration = span.duration || spanEndTime - spanStartTime
|
||||
const startOffset = spanStartTime - parentStartTime // Time from parent start to this span's start
|
||||
|
||||
// Calculate the position relative to the workflow start time (for Gantt chart style)
|
||||
const relativeStartPercent =
|
||||
totalDuration > 0 ? ((spanStartTime - workflowStartTime) / totalDuration) * 100 : 0
|
||||
const durationPercent = totalDuration > 0 ? (duration / totalDuration) * 100 : 0
|
||||
|
||||
// Ensure values are within valid range
|
||||
const safeStartPercent = Math.min(100, Math.max(0, relativeStartPercent))
|
||||
const safeWidthPercent = Math.max(2, Math.min(100 - safeStartPercent, durationPercent))
|
||||
|
||||
// For parent-relative timing display
|
||||
const startOffsetPercentage = totalDuration > 0 ? (startOffset / totalDuration) * 100 : 0
|
||||
|
||||
// Get appropriate icon based on span type
|
||||
const getSpanIcon = () => {
|
||||
const type = span.type.toLowerCase()
|
||||
|
||||
// Expand/collapse for spans with children
|
||||
if (hasChildren || hasToolCalls) {
|
||||
return expanded ? <ChevronDown className="h-4 w-4" /> : <ChevronRight className="h-4 w-4" />
|
||||
}
|
||||
|
||||
// Block type specific icons
|
||||
if (type === 'agent') {
|
||||
return <AgentIcon className="h-3 w-3 text-[#7F2FFF]" />
|
||||
}
|
||||
|
||||
if (type === 'evaluator') {
|
||||
return <ChartBarIcon className="h-3 w-3 text-[#2FA1FF]" />
|
||||
}
|
||||
|
||||
if (type === 'condition') {
|
||||
return <ConditionalIcon className="h-3 w-3 text-[#FF972F]" />
|
||||
}
|
||||
|
||||
if (type === 'router') {
|
||||
return <ConnectIcon className="h-3 w-3 text-[#2FA1FF]" />
|
||||
}
|
||||
|
||||
if (type === 'model') {
|
||||
return <Cpu className="h-3 w-3 text-[#10a37f]" />
|
||||
}
|
||||
|
||||
if (type === 'function') {
|
||||
return <CodeIcon className="h-3 w-3 text-[#FF402F]" />
|
||||
}
|
||||
|
||||
if (type === 'tool') {
|
||||
return <ExternalLink className="h-3 w-3 text-[#f97316]" />
|
||||
}
|
||||
|
||||
if (type === 'api') {
|
||||
return <ApiIcon className="h-3 w-3 text-[#2F55FF]" />
|
||||
}
|
||||
|
||||
return <Code className="h-3 w-3 text-muted-foreground" />
|
||||
}
|
||||
|
||||
// Format milliseconds as +XXms for relative timing
|
||||
const formatRelativeTime = (ms: number) => {
|
||||
if (ms === 0) return 'start'
|
||||
return `+${ms}ms`
|
||||
}
|
||||
|
||||
// Get color based on span type
|
||||
const getSpanColor = (type: string) => {
|
||||
switch (type.toLowerCase()) {
|
||||
case 'agent':
|
||||
return '#7F2FFF' // Purple from AgentBlock
|
||||
case 'provider':
|
||||
return '#818cf8' // Indigo for provider
|
||||
case 'model':
|
||||
return '#10a37f' // Green from OpenAIBlock
|
||||
case 'function':
|
||||
return '#FF402F' // Orange-red from FunctionBlock
|
||||
case 'tool':
|
||||
return '#f97316' // Orange for tools
|
||||
case 'router':
|
||||
return '#2FA1FF' // Blue from RouterBlock
|
||||
case 'condition':
|
||||
return '#FF972F' // Orange from ConditionBlock
|
||||
case 'evaluator':
|
||||
return '#2FA1FF' // Blue from EvaluatorBlock
|
||||
case 'api':
|
||||
return '#2F55FF' // Blue from ApiBlock
|
||||
default:
|
||||
return '#6b7280' // Gray for others
|
||||
}
|
||||
}
|
||||
|
||||
const spanColor = getSpanColor(span.type)
|
||||
|
||||
// Format duration to be more readable
|
||||
const formatDuration = (ms: number) => {
|
||||
if (ms < 1000) return `${ms}ms`
|
||||
return `${(ms / 1000).toFixed(2)}s`
|
||||
}
|
||||
|
||||
// Extract model name from span name using a more general pattern
|
||||
const extractModelName = (spanName: string) => {
|
||||
// Try to match model name in parentheses
|
||||
const modelMatch = spanName.match(/\(([\w\.-]+)\)/i)
|
||||
return modelMatch ? modelMatch[1] : ''
|
||||
}
|
||||
|
||||
// Format span name for display
|
||||
const formatSpanName = (span: TraceSpan) => {
|
||||
if (span.type === 'model') {
|
||||
const modelName = extractModelName(span.name)
|
||||
|
||||
if (span.name.includes('Initial response')) {
|
||||
return (
|
||||
<>
|
||||
Initial response{' '}
|
||||
{modelName && <span className="text-xs opacity-75">({modelName})</span>}
|
||||
</>
|
||||
)
|
||||
} else if (span.name.includes('(iteration')) {
|
||||
const iterMatch = span.name.match(/\(iteration (\d+)\)/)
|
||||
const iterNum = iterMatch ? iterMatch[1] : ''
|
||||
return (
|
||||
<>
|
||||
Model response{' '}
|
||||
{iterNum && <span className="text-xs opacity-75">(iteration {iterNum})</span>}{' '}
|
||||
{modelName && <span className="text-xs opacity-75">({modelName})</span>}
|
||||
</>
|
||||
)
|
||||
} else if (span.name.includes('Model Generation')) {
|
||||
return (
|
||||
<>
|
||||
Model Generation{' '}
|
||||
{modelName && <span className="text-xs opacity-75">({modelName})</span>}
|
||||
</>
|
||||
)
|
||||
}
|
||||
}
|
||||
|
||||
return span.name
|
||||
}
|
||||
|
||||
return (
|
||||
<div
|
||||
className={cn(
|
||||
'border-b last:border-b-0 transition-colors',
|
||||
expanded ? 'bg-secondary/10' : 'hover:bg-secondary/5'
|
||||
)}
|
||||
>
|
||||
{/* Span header */}
|
||||
<div
|
||||
className="flex items-center py-1.5 px-2 cursor-pointer"
|
||||
onClick={() => setExpanded(!expanded)}
|
||||
style={{ paddingLeft: `${depth * 16 + 8}px` }}
|
||||
>
|
||||
<div className="mr-2 flex items-center justify-center w-5">{getSpanIcon()}</div>
|
||||
|
||||
<div className="flex-1 flex items-center min-w-0">
|
||||
<div className="flex flex-col min-w-0 mr-3">
|
||||
<div className="flex items-center space-x-2">
|
||||
<span
|
||||
className={cn(
|
||||
'text-sm font-medium truncate',
|
||||
span.status === 'error' && 'text-red-500'
|
||||
)}
|
||||
>
|
||||
{formatSpanName(span)}
|
||||
</span>
|
||||
{depth > 0 && (
|
||||
<span className="text-xs text-muted-foreground whitespace-nowrap">
|
||||
{span.relativeStartMs !== undefined
|
||||
? `+${span.relativeStartMs}ms`
|
||||
: formatRelativeTime(startOffset)}
|
||||
</span>
|
||||
)}
|
||||
{span.tokens && (
|
||||
<span className="text-xs bg-secondary/40 px-1.5 py-0.5 rounded text-muted-foreground whitespace-nowrap">
|
||||
{span.tokens} tokens
|
||||
</span>
|
||||
)}
|
||||
</div>
|
||||
<span className="text-xs text-muted-foreground">{formatDuration(duration)}</span>
|
||||
</div>
|
||||
|
||||
<div className="ml-auto flex items-center gap-3 shrink-0">
|
||||
{/* Timeline visualization - all spans are positioned relative to workflow start */}
|
||||
<div className="w-48 h-2 bg-secondary/20 rounded-full overflow-hidden relative">
|
||||
<div
|
||||
className="h-full rounded-full absolute"
|
||||
style={{
|
||||
left: `${safeStartPercent}%`,
|
||||
width: `${safeWidthPercent}%`,
|
||||
backgroundColor: spanColor,
|
||||
}}
|
||||
/>
|
||||
</div>
|
||||
|
||||
{/* Duration text - always show in ms */}
|
||||
<span className="text-xs text-muted-foreground w-20 text-right font-mono tabular-nums">
|
||||
{`${duration}ms`}
|
||||
</span>
|
||||
</div>
|
||||
</div>
|
||||
</div>
|
||||
|
||||
{/* Children and tool calls */}
|
||||
{expanded && (
|
||||
<div>
|
||||
{/* Render child spans */}
|
||||
{hasChildren && (
|
||||
<div>
|
||||
{span.children!.map((childSpan, index) => (
|
||||
<TraceSpanItem
|
||||
key={index}
|
||||
span={childSpan}
|
||||
depth={depth + 1}
|
||||
totalDuration={totalDuration}
|
||||
isLast={index === span.children!.length - 1}
|
||||
parentStartTime={spanStartTime}
|
||||
workflowStartTime={workflowStartTime}
|
||||
/>
|
||||
))}
|
||||
</div>
|
||||
)}
|
||||
|
||||
{/* Render tool calls as spans */}
|
||||
{hasToolCalls && (
|
||||
<div>
|
||||
{span.toolCalls!.map((toolCall, index) => {
|
||||
// Create a pseudo-span for each tool call
|
||||
const toolStartTime = toolCall.startTime
|
||||
? new Date(toolCall.startTime).getTime()
|
||||
: spanStartTime
|
||||
const toolEndTime = toolCall.endTime
|
||||
? new Date(toolCall.endTime).getTime()
|
||||
: toolStartTime + (toolCall.duration || 0)
|
||||
|
||||
const toolSpan: TraceSpan = {
|
||||
id: `${span.id}-tool-${index}`,
|
||||
name: toolCall.name,
|
||||
type: 'tool',
|
||||
duration: toolCall.duration || toolEndTime - toolStartTime,
|
||||
startTime: new Date(toolStartTime).toISOString(),
|
||||
endTime: new Date(toolEndTime).toISOString(),
|
||||
status: toolCall.error ? 'error' : 'success',
|
||||
}
|
||||
|
||||
return (
|
||||
<TraceSpanItem
|
||||
key={`tool-${index}`}
|
||||
span={toolSpan}
|
||||
depth={depth + 1}
|
||||
totalDuration={totalDuration}
|
||||
isLast={index === span.toolCalls!.length - 1}
|
||||
parentStartTime={spanStartTime}
|
||||
workflowStartTime={workflowStartTime}
|
||||
/>
|
||||
)
|
||||
})}
|
||||
</div>
|
||||
)}
|
||||
</div>
|
||||
)}
|
||||
</div>
|
||||
)
|
||||
}
|
||||
@@ -22,6 +22,20 @@ export interface ToolCallMetadata {
|
||||
toolCalls?: ToolCall[]
|
||||
}
|
||||
|
||||
export interface TraceSpan {
|
||||
id: string
|
||||
name: string
|
||||
type: string
|
||||
duration: number // in milliseconds
|
||||
startTime: string
|
||||
endTime: string
|
||||
children?: TraceSpan[]
|
||||
toolCalls?: ToolCall[]
|
||||
status?: 'success' | 'error'
|
||||
tokens?: number
|
||||
relativeStartMs?: number // Time in ms from the start of the parent span
|
||||
}
|
||||
|
||||
export interface WorkflowLog {
|
||||
id: string
|
||||
workflowId: string
|
||||
@@ -32,7 +46,10 @@ export interface WorkflowLog {
|
||||
trigger: string | null
|
||||
createdAt: string
|
||||
workflow?: WorkflowData | null
|
||||
metadata?: ToolCallMetadata | Record<string, any> // Add metadata for tool calls
|
||||
metadata?: ToolCallMetadata & {
|
||||
traceSpans?: TraceSpan[]
|
||||
totalDuration?: number
|
||||
}
|
||||
}
|
||||
|
||||
export interface LogsResponse {
|
||||
|
||||
Reference in New Issue
Block a user