Merge pull request #1664 from liam-hq/add-mastra-to-chat

FunamaYukina · web-flow · commit 848fc71c9e69 · 2025-05-16T06:01:02.000Z
🔧 Remove `langchain` and add `vercel-ai-sdk` &amp; `mastra`
diff --git a/config/dependency_decisions.yml b/config/dependency_decisions.yml
@@ -139,3 +139,9 @@
     :why: Compatible with Apache-2.0 license. See https://opensource.org/license/artistic-2-0
     :versions: []
     :when: 2025-05-08 05:24:20.998252000 Z
+- - :approve
+  - "@mastra/core"
+  - :who: Liam @FunamaYukina
+    :why: The license is Elastic-2.0. This is not an OSS license, so to be on the safe side, we'll approve it on a package-by-package basis.
+    :versions: []
+    :when: 2025-05-13 00:30:00.632489000 Z
diff --git a/frontend/apps/app/.gitignore b/frontend/apps/app/.gitignore
@@ -52,3 +52,6 @@ prism.wasm
 !.env.local
 
 *storybook.log
+
+# Mastra
+.mastra
diff --git a/frontend/apps/app/app/api/chat/route.ts b/frontend/apps/app/app/api/chat/route.ts
@@ -1,9 +1,6 @@
-import { langfuseHandler } from '@/libs/langfuse/langfuseHandler'
-import {} from '@langchain/core/messages'
-import { ChatPromptTemplate } from '@langchain/core/prompts'
-import { ChatOpenAI } from '@langchain/openai'
+import { mastra } from '@/lib/mastra'
 import type { Schema, TableGroup } from '@liam-hq/db-structure'
-import { Document } from 'langchain/document'
+import * as Sentry from '@sentry/nextjs'
 import { NextResponse } from 'next/server'
 
 // Export TableGroupData type for compatibility
@@ -13,7 +10,7 @@ export type TableGroupData = TableGroup
 const tableToDocument = (
   tableName: string,
   tableData: Schema['tables'][string],
-): Document => {
+): string => {
   // Table description
   const tableDescription = `Table: ${tableName}\nDescription: ${tableData.comment || 'No description'}\n`
 
@@ -39,30 +36,20 @@ const tableToDocument = (
   }
 
   // Combine all information
-  const tableText = `${tableDescription}${columnsText}${primaryKeyText}`
-
-  return new Document({
-    pageContent: tableText,
-    metadata: { tableName },
-  })
+  return `${tableDescription}${columnsText}${primaryKeyText}`
 }
 
 // Convert relationship data to text document
 const relationshipToDocument = (
   relationshipName: string,
   relationshipData: Schema['relationships'][string],
-): Document => {
-  const relationshipText = `Relationship: ${relationshipName}
+): string => {
+  return `Relationship: ${relationshipName}
 From Table: ${relationshipData.primaryTableName}
 From Column: ${relationshipData.primaryColumnName}
 To Table: ${relationshipData.foreignTableName}
 To Column: ${relationshipData.foreignColumnName}
 Type: ${relationshipData.cardinality || 'unknown'}\n`
-
-  return new Document({
-    pageContent: relationshipText,
-    metadata: { relationshipName },
-  })
 }
 
 // Convert table groups to text document
@@ -99,7 +86,7 @@ const convertSchemaToText = (schema: Schema): string => {
     schemaText += 'TABLES:\n\n'
     for (const [tableName, tableData] of Object.entries(schema.tables)) {
       const tableDoc = tableToDocument(tableName, tableData)
-      schemaText = `${schemaText}${tableDoc.pageContent}\n\n`
+      schemaText = `${schemaText}${tableDoc}\n\n`
     }
   }
 
@@ -113,7 +100,7 @@ const convertSchemaToText = (schema: Schema): string => {
         relationshipName,
         relationshipData,
       )
-      schemaText = `${schemaText}${relationshipDoc.pageContent}\n\n`
+      schemaText = `${schemaText}${relationshipDoc}\n\n`
     }
   }
 
@@ -141,7 +128,7 @@ export async function POST(request: Request) {
     )
   }
 
-  // Format chat history for prompt template
+  // Format chat history for prompt
   const formattedChatHistory =
     // eslint-disable-next-line @typescript-eslint/no-unsafe-member-access
     history && history.length > 0
@@ -155,126 +142,41 @@ export async function POST(request: Request) {
   // Convert schema to text
   const schemaText = convertSchemaToText(schemaData)
 
-  // Create a streaming model
-  const streamingModel = new ChatOpenAI({
-    modelName: 'o4-mini-2025-04-16',
-    streaming: true,
-    callbacks: [langfuseHandler],
-  })
-
-  // Create a prompt template with full schema context and chat history
-  const prompt = ChatPromptTemplate.fromTemplate(`
-You are a database schema expert.
-Answer questions about the user's schema and provide advice on database design.
-Follow these guidelines:
-
-1. Clearly explain the structure of the schema, tables, and relationships.
-2. Provide advice based on good database design principles.
-3. Share best practices for normalization, indexing, and performance.
-4. When using technical terms, include brief explanations.
-5. Provide only information directly related to the question, avoiding unnecessary details.
-6. Format your responses using GitHub Flavored Markdown (GFM) for better readability.
-
-Your goal is to help users understand and optimize their database schemas.
+  try {
+    // Get the agent from Mastra
+    const agent = mastra.getAgent('databaseSchemaAgent')
+    if (!agent) {
+      throw new Error('databaseSchemaAgent not found in Mastra instance')
+    }
 
+    // Create a response using the agent
+    const response = await agent.generate([
+      {
+        role: 'system',
+        content: `
 Complete Schema Information:
 ${schemaText}
 
 Previous conversation:
-{chat_history}
-
-Question: {input}
-
-Based on the schema information provided and considering any previous conversation, answer the question thoroughly and accurately.
-`)
-
-  // Create streaming chain
-  const streamingChain = prompt.pipe(streamingModel)
-
-  // Generate streaming response
-  const stream = await streamingChain.stream(
-    {
-      input: message,
-      chat_history: formattedChatHistory,
-    },
-    {
-      callbacks: [langfuseHandler],
-      metadata: {
-        endpoint: '/api/chat',
-
-        method: 'POST',
-        messageLength: message.length,
-        // eslint-disable-next-line @typescript-eslint/no-unsafe-member-access
-        hasHistory: history ? history.length > 0 : false,
+${formattedChatHistory}
+`,
       },
-    },
-  )
-
-  // Create a TransformStream to convert the LangChain stream to a ReadableStream
-  const encoder = new TextEncoder()
-  const { readable, writable } = new TransformStream()
-  const writer = writable.getWriter()
-
-  // Define types for content processing
-  type ContentItem = string | { type: string; text: string } | unknown
-
-  // Extract content processing to a separate function
-  const extractTextContent = (
-    content: string | ContentItem[] | unknown,
-  ): string => {
-    if (typeof content === 'string') {
-      return content
-    }
-
-    if (!Array.isArray(content)) {
-      return ''
-    }
-
-    // Process array content
-    return content.reduce((text, item) => {
-      if (typeof item === 'string') {
-        return text + item
-      }
-
-      if (
-        item &&
-        typeof item === 'object' &&
-        'type' in item &&
-        // eslint-disable-next-line @typescript-eslint/no-unsafe-member-access
-        item.type === 'text' &&
-        'text' in item &&
-        // eslint-disable-next-line @typescript-eslint/no-unsafe-member-access
-        typeof item.text === 'string'
-      ) {
-        // eslint-disable-next-line @typescript-eslint/no-unsafe-member-access
-        return text + item.text
-      }
-
-      return text
-    }, '')
-  }
+      {
+        role: 'user',
+        content: message,
+      },
+    ])
 
-  // Main stream processing function - simplified
-  const processStream = async () => {
-    try {
-      for await (const chunk of stream) {
-        const textContent = extractTextContent(chunk.content)
-        await writer.write(encoder.encode(textContent))
-      }
-    } catch (error) {
-      console.error('Error processing stream:', error)
-    } finally {
-      await writer.close()
-    }
+    return new Response(response.text, {
+      headers: {
+        'Content-Type': 'text/plain; charset=utf-8',
+      },
+    })
+  } catch (error) {
+    Sentry.captureException(error)
+    return NextResponse.json(
+      { error: 'Failed to generate response' },
+      { status: 500 },
+    )
   }
-
-  // Execute the processing function
-  processStream()
-
-  // Return the streaming response
-  return new Response(readable, {
-    headers: {
-      'Content-Type': 'text/plain; charset=utf-8',
-    },
-  })
 }
diff --git a/frontend/apps/app/instrumentation.ts b/frontend/apps/app/instrumentation.ts
@@ -1,7 +1,19 @@
 import { validateConfig } from '@liam-hq/github'
 import * as Sentry from '@sentry/nextjs'
+import { registerOTel } from '@vercel/otel'
+import { LangfuseExporter } from 'langfuse-vercel'
 
 export async function register() {
+  registerOTel({
+    serviceName: 'liam-app',
+    traceExporter: new LangfuseExporter({
+      publicKey: process.env.LANGFUSE_PUBLIC_KEY || '',
+      secretKey: process.env.LANGFUSE_SECRET_KEY || '',
+      baseUrl: process.env.LANGFUSE_BASE_URL || 'https://cloud.langfuse.com',
+      environment: process.env.NEXT_PUBLIC_ENV_NAME || 'development',
+    }),
+  })
+
   if (process.env.NEXT_RUNTIME === 'nodejs') {
     await import('./sentry.server.config')
   }
diff --git a/frontend/apps/app/lib/mastra/agents/databaseSchemaAgent.ts b/frontend/apps/app/lib/mastra/agents/databaseSchemaAgent.ts
@@ -0,0 +1,26 @@
+import { openai } from '@ai-sdk/openai'
+import type { Metric } from '@mastra/core'
+import { Agent, type ToolsInput } from '@mastra/core/agent'
+
+export const databaseSchemaAgent: Agent<
+  'Database Schema Expert',
+  ToolsInput,
+  Record<string, Metric>
+> = new Agent({
+  name: 'Database Schema Expert',
+  instructions: `
+You are a database schema expert.
+Answer questions about the user's schema and provide advice on database design.
+Follow these guidelines:
+
+1. Clearly explain the structure of the schema, tables, and relationships.
+2. Provide advice based on good database design principles.
+3. Share best practices for normalization, indexing, and performance.
+4. When using technical terms, include brief explanations.
+5. Provide only information directly related to the question, avoiding unnecessary details.
+6. Format your responses using GitHub Flavored Markdown (GFM) for better readability.
+
+Your goal is to help users understand and optimize their database schemas.
+`,
+  model: openai('o4-mini-2025-04-16'),
+})
diff --git a/frontend/apps/app/lib/mastra/index.ts b/frontend/apps/app/lib/mastra/index.ts
@@ -0,0 +1,29 @@
+import { Mastra } from '@mastra/core'
+import { LangfuseExporter } from 'langfuse-vercel'
+import { databaseSchemaAgent } from './agents/databaseSchemaAgent'
+/**
+ * Mastra instance with Langfuse tracing integration
+ *
+ * The telemetry configuration is set up to work with the Langfuse exporter
+ * configured in instrumentation.ts. The serviceName must be 'ai' to match
+ * the ATTR_SERVICE_NAME in the NodeSDK configuration.
+ */
+export const mastra = new Mastra({
+  agents: {
+    databaseSchemaAgent,
+  },
+  telemetry: {
+    serviceName: 'ai', // Must match ATTR_SERVICE_NAME in instrumentation.ts
+    enabled: true,
+    export: {
+      type: 'custom',
+      exporter: new LangfuseExporter({
+        publicKey: process.env.LANGFUSE_PUBLIC_KEY || '',
+        secretKey: process.env.LANGFUSE_SECRET_KEY || '',
+        baseUrl: process.env.LANGFUSE_BASE_URL || 'https://cloud.langfuse.com',
+        environment: process.env.NEXT_PUBLIC_ENV_NAME || 'development',
+        debug: true,
+      }),
+    },
+  },
+})
diff --git a/frontend/apps/app/libs/langfuse/langfuseHandler.ts b/frontend/apps/app/libs/langfuse/langfuseHandler.ts
diff --git a/frontend/apps/app/next.config.ts b/frontend/apps/app/next.config.ts
@@ -13,6 +13,9 @@ if (process.env.VERCEL_ENV === 'production') {
 }
 
 const nextConfig: NextConfig = {
+  // Server-only packages that should not be bundled on the client
+  serverExternalPackages: ['@mastra/*'],
+
   // NOTE: Exclude Prisma-related packages from the bundle
   // These packages are installed separately in the node_modules/@prisma directory
   // Excluding them prevents `Error: Cannot find module 'fs'` errors in the build process
diff --git a/frontend/apps/app/package.json b/frontend/apps/app/package.json
@@ -3,6 +3,7 @@
   "private": true,
   "version": "0.1.0",
   "dependencies": {
+    "@ai-sdk/openai": "1.3.22",
     "@codemirror/commands": "6.8.1",
     "@codemirror/lang-json": "6.0.1",
     "@codemirror/lang-yaml": "6.1.2",
@@ -11,15 +12,14 @@
     "@codemirror/merge": "6.10.0",
     "@codemirror/state": "6.5.2",
     "@codemirror/view": "6.36.6",
-    "@langchain/core": "0.3.49",
-    "@langchain/openai": "0.5.7",
     "@lezer/highlight": "1.2.1",
     "@liam-hq/db": "workspace:*",
     "@liam-hq/db-structure": "workspace:*",
     "@liam-hq/erd-core": "workspace:*",
     "@liam-hq/github": "workspace:*",
     "@liam-hq/jobs": "workspace:*",
     "@liam-hq/ui": "workspace:*",
+    "@mastra/core": "0.9.3",
     "@next/third-parties": "15.3.1",
     "@octokit/auth-app": "7.2.1",
     "@octokit/rest": "21.1.1",
@@ -28,16 +28,17 @@
     "@trigger.dev/build": "3.3.17",
     "@trigger.dev/sdk": "3.3.17",
     "@types/react-syntax-highlighter": "15.5.13",
+    "@vercel/otel": "1.12.0",
+    "ai": "4.3.15",
     "cheerio": "1.0.0",
     "clsx": "2.1.1",
     "codemirror": "6.0.1",
     "date-fns": "4.1.0",
     "diff": "7.0.0",
     "dotenv": "16.5.0",
     "flags": "4.0.0",
-    "langchain": "0.3.24",
     "langfuse": "3.37.2",
-    "langfuse-langchain": "3.37.2",
+    "langfuse-vercel": "3.37.2",
     "lucide-react": "0.503.0",
     "next": "15.3.1",
     "react": "18.3.1",
diff --git a/frontend/apps/app/tsconfig.json b/frontend/apps/app/tsconfig.json
diff --git a/pnpm-lock.yaml b/pnpm-lock.yaml

Original file line number	Diff line number	Diff line change
`@@ -13,6 +13,9 @@ if (process.env.VERCEL_ENV === 'production') {`
`13`	`13`	`}`
`14`	`14`
`15`	`15`	`const nextConfig: NextConfig = {`
	`16`	`+ // Server-only packages that should not be bundled on the client`
	`17`	`+ serverExternalPackages: ['@mastra/*'],`
	`18`	`+`
`16`	`19`	`// NOTE: Exclude Prisma-related packages from the bundle`
`17`	`20`	`// These packages are installed separately in the node_modules/@prisma directory`
`18`	`21`	// Excluding them prevents `Error: Cannot find module 'fs'` errors in the build process