Jakob Wennberg
134 lines
3.8 KiB
TypeScript
134 lines
3.8 KiB
TypeScript
import { ChatAnthropic } from '@langchain/anthropic'
|
|
import { createReactAgent } from '@langchain/langgraph/prebuilt'
|
|
import { HumanMessage, AIMessage } from '@langchain/core/messages'
|
|
import type { StructuredToolInterface } from '@langchain/core/tools'
|
|
import { CHATBOT_CONFIG } from './config'
|
|
import {
|
|
SYSTEM_PROMPT_DATA,
|
|
SYSTEM_PROMPT_HYBRID,
|
|
formatConversationHistory,
|
|
} from './prompts'
|
|
import type { ChatMessage } from '@/types/chat'
|
|
import type { RouteType } from './router'
|
|
|
|
export interface AgentStreamEvent {
|
|
type: 'tool_start' | 'content' | 'done'
|
|
toolName?: string
|
|
content?: string
|
|
toolResults?: ToolResultEntry[]
|
|
}
|
|
|
|
export interface ToolResultEntry {
|
|
toolName: string
|
|
result: string
|
|
}
|
|
|
|
/**
|
|
* Run the LangGraph agent with tool calling and stream events.
|
|
*/
|
|
export async function* streamAgentResponse(options: {
|
|
query: string
|
|
route: RouteType
|
|
tools: StructuredToolInterface[]
|
|
conversationHistory: ChatMessage[]
|
|
ragContext?: string
|
|
}): AsyncGenerator<AgentStreamEvent> {
|
|
const { query, route, tools, conversationHistory, ragContext } = options
|
|
|
|
// Build system prompt based on route
|
|
const historyText = formatConversationHistory(
|
|
conversationHistory.slice(-CHATBOT_CONFIG.maxHistoryMessages).map((m) => ({
|
|
role: m.role,
|
|
content: m.content,
|
|
}))
|
|
)
|
|
|
|
let systemPrompt: string
|
|
if (route === 'data') {
|
|
systemPrompt = SYSTEM_PROMPT_DATA.replace('{history}', historyText)
|
|
} else {
|
|
const context = ragContext || 'Ingen specifik kontext hittades i kunskapsbasen.'
|
|
systemPrompt = SYSTEM_PROMPT_HYBRID
|
|
.replace('{context}', context)
|
|
.replace('{history}', historyText)
|
|
}
|
|
|
|
// Create the model
|
|
const model = new ChatAnthropic({
|
|
modelName: CHATBOT_CONFIG.agentModel,
|
|
maxTokens: CHATBOT_CONFIG.agentMaxTokens,
|
|
temperature: CHATBOT_CONFIG.temperature,
|
|
anthropicApiKey: process.env.ANTHROPIC_API_KEY,
|
|
})
|
|
|
|
// Create the agent
|
|
const agent = createReactAgent({
|
|
llm: model,
|
|
tools,
|
|
prompt: systemPrompt,
|
|
})
|
|
|
|
// Build input messages
|
|
const messages: (HumanMessage | AIMessage)[] = []
|
|
|
|
// Add recent history as messages for the agent
|
|
const recent = conversationHistory.slice(-CHATBOT_CONFIG.maxHistoryMessages)
|
|
for (const msg of recent) {
|
|
if (msg.role === 'user') {
|
|
messages.push(new HumanMessage(msg.content))
|
|
} else {
|
|
messages.push(new AIMessage(msg.content))
|
|
}
|
|
}
|
|
messages.push(new HumanMessage(query))
|
|
|
|
// Track tool results for artifact generation
|
|
const toolResults: ToolResultEntry[] = []
|
|
|
|
// Stream the agent execution using streamEvents for fine-grained control
|
|
const eventStream = agent.streamEvents(
|
|
{ messages },
|
|
{
|
|
version: 'v2',
|
|
recursionLimit: CHATBOT_CONFIG.maxAgentIterations * 2 + 1,
|
|
}
|
|
)
|
|
|
|
for await (const event of eventStream) {
|
|
// Tool start events
|
|
if (event.event === 'on_tool_start') {
|
|
yield { type: 'tool_start', toolName: event.name }
|
|
}
|
|
|
|
// Tool end events — capture results
|
|
if (event.event === 'on_tool_end') {
|
|
const output = event.data?.output
|
|
const result = typeof output === 'string' ? output : JSON.stringify(output ?? '')
|
|
toolResults.push({
|
|
toolName: event.name,
|
|
result,
|
|
})
|
|
}
|
|
|
|
// LLM streaming tokens (final response text)
|
|
if (event.event === 'on_chat_model_stream') {
|
|
const chunk = event.data?.chunk
|
|
if (chunk) {
|
|
const content = typeof chunk.content === 'string'
|
|
? chunk.content
|
|
: Array.isArray(chunk.content)
|
|
? chunk.content
|
|
.filter((c: { type: string }) => c.type === 'text')
|
|
.map((c: { text: string }) => c.text)
|
|
.join('')
|
|
: ''
|
|
if (content) {
|
|
yield { type: 'content', content }
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
yield { type: 'done', toolResults }
|
|
}
|