feat(core): support chroma

packages/core/package.json

@@ -132,13 +132,17 @@
   },
   "peerDependencies": {
     "@supabase/supabase-js": "^2.26.0",
+    "chromadb": "^1.5.3",
     "langchain": "^0.0.96",
     "ofetch": "^1.1.1"
   },
   "peerDependenciesMeta": {
     "@supabase/supabase-js": {
       "optional": true
     },
+    "chromadb": {
+      "optional": true
+    },
     "langchain": {
       "optional": true
     },

packages/core/src/storage/chroma/index.ts

@@ -0,0 +1,64 @@
+import { Chroma } from 'langchain/vectorstores/chroma'
+import { Document } from 'langchain/document'
+import { RecursiveCharacterTextSplitter } from 'langchain/text_splitter'
+import type { WebInfoData } from '../../types'
+import { VectorStorage } from '../types'
+import type { StorageType, VectorConfig } from '../types'
+
+export interface ChromaVectorConfig extends VectorConfig {
+  url: string
+}
+export class ChromaVectorStorage extends VectorStorage<ChromaVectorConfig> {
+  constructor(config: ChromaVectorConfig, data?: WebInfoData) {
+    super(config, data)
+  }
+
+  async save(data?: WebInfoData) {
+    if (!data && !this.data)
+      throw new Error('VectorStorage error: No Storage Data')
+
+    const storageData = (data || this.data)!
+    const rawDoc = new Document({ pageContent: storageData.content, metadata: this.config.metaData })
+
+    /* Split text into chunks */
+    const textSplitter = new RecursiveCharacterTextSplitter({
+      chunkSize: 300,
+      chunkOverlap: 40,
+    })
+
+    const docs = await textSplitter.splitDocuments([rawDoc])
+    await Chroma.fromDocuments(docs, this.config.embeddingsInfo.embeddings, {
+      collectionName: this.config.embeddingsInfo.indexName,
+      url: this.config.url,
+    })
+  }
+
+  async getRetriever() {
+    const vectorStore = await Chroma.fromExistingCollection(
+      this.config.embeddingsInfo.embeddings,
+      {
+        collectionName: this.config.embeddingsInfo.indexName,
+        url: this.config.url,
+        // TODO: remove this eslint rule after langchainjs add filter options
+        // eslint-disable-next-line @typescript-eslint/ban-ts-comment
+        // @ts-expect-error
+        filter: {
+          appName: this.config.metaData.appName,
+          botId: this.config.metaData.botId,
+          userId: this.config.metaData.userId,
+        },
+      })
+    return vectorStore.asRetriever()
+  }
+
+  getConfig(): ChromaVectorConfig {
+    return this.config
+  }
+
+  getType(): StorageType {
+    return {
+      type: 'VectorStorage',
+      name: 'ChromaVectorConfig',
+    }
+  }
+}

packages/core/src/storage/index.ts

@@ -1,5 +1,6 @@
 import { NotionDataStorage } from './notion'
 import { SupabaseImageStorage, SupabaseVectorStorage } from './supabase'
+import { ChromaVectorStorage } from './chroma'
 import type { TStorage } from './types'
 
 export * from './types'
@@ -20,5 +21,6 @@ export const storageInfo: StorageInfo = {
   },
   VectorStorage: {
     SupabaseVectorStorage,
+    ChromaVectorStorage,
   },
 }

packages/core/vite.config.ts

@@ -42,13 +42,15 @@ export default defineConfig({
       external: [
         '@supabase/supabase-js',
         /langchain\/.*/,
+        'chromadb',
         'ofetch',
       ],
       output: {
         // Provide global variables to use in the UMD build
         // for externalized deps
         globals: {
           '@supabase/supabase-js': '@supabase/supabase-js',
+          'chromadb': 'chromadb',
           'ofetch': 'ofetch',
         },
       },

server/nuxt3/composables/config.ts

@@ -342,6 +342,17 @@ export const vectorStorageConfig: BasicConfig<ModelsConfig> = {
       },
       output: '',
     },
+    ChromaVectorStorage: {
+      displayName: 'Chroma Storage',
+      config: {
+        url: {
+          label: 'Chroma URL',
+          value: '',
+          require: true,
+        },
+      },
+      output: '',
+    },
   },
 }
 export const llmConfig: BasicConfig<ModelsConfig> = {

server/nuxt3/package.json

@@ -29,6 +29,7 @@
     "@vue-flow/controls": "^1.1.0",
     "@vue-flow/core": "^1.20.2",
     "@vue-flow/minimap": "^1.1.1",
+    "chromadb": "^1.5.3",
     "langchain": "^0.0.96",
     "satori": "^0.10.1",
     "satori-html": "^0.3.2"