feat(RAG): [wip] RAG capabilities

Dockerfile

@@ -1,7 +1,7 @@
 FROM node:22.16.0-alpine3.22 AS base
 
-# Install bash & curl for entrypoint script compatibility
+# Install bash & curl for entrypoint script compatibility, graphicsmagick for pdf2pic, and vips-dev & build-base for sharp 
-RUN apk add --no-cache bash curl
+RUN apk add --no-cache bash curl graphicsmagick vips-dev build-base
 
 # All deps stage
 FROM base AS deps

admin/app/controllers/rag_controller.ts

@@ -0,0 +1,18 @@
+import { cuid } from '@adonisjs/core/helpers'
+import type { HttpContext } from '@adonisjs/core/http'
+import app from '@adonisjs/core/services/app'
+
+export default class RagsController {
+  public async upload({ request, response }: HttpContext) {
+    const uploadedFile = request.file('file')
+    if (!uploadedFile) {
+      return response.status(400).json({ error: 'No file uploaded' })
+    }
+
+    const fileName = `${cuid()}.${uploadedFile.extname}`
+
+    await uploadedFile.move(app.makePath('storage/uploads'), {
+      name: fileName,
+    })
+  }
+}

admin/app/models/kv_store.ts

@@ -0,0 +1,46 @@
+import { DateTime } from 'luxon'
+import { BaseModel, column, SnakeCaseNamingStrategy } from '@adonisjs/lucid/orm'
+import type { KVStoreKey, KVStoreValue } from '../../types/kv_store.js'
+
+/**
+ * Generic key-value store model for storing various settings
+ * that don't necessitate their own dedicated models.
+ */
+export default class KVStore extends BaseModel {
+  static namingStrategy = new SnakeCaseNamingStrategy()
+
+  @column({ isPrimary: true })
+  declare id: number
+
+  @column()
+  declare key: KVStoreKey
+
+  @column()
+  declare value: KVStoreValue
+
+  @column.dateTime({ autoCreate: true })
+  declare created_at: DateTime
+
+  @column.dateTime({ autoCreate: true, autoUpdate: true })
+  declare updated_at: DateTime
+
+  /**
+   * Get a setting value by key
+   */
+  static async getValue(key: KVStoreKey): Promise<KVStoreValue> {
+    const setting = await this.findBy('key', key)
+    return setting?.value ?? null
+  }
+
+  /**
+   * Set a setting value by key (creates if not exists)
+   */
+  static async setValue(key: KVStoreKey, value: KVStoreValue): Promise<KVStore> {
+    const setting = await this.firstOrCreate({ key }, { key, value })
+    if (setting.value !== value) {
+      setting.value = value
+      await setting.save()
+    }
+    return setting
+  }
+}

admin/app/services/benchmark_service.ts

@@ -426,21 +426,10 @@ export class BenchmarkService {
     }
 
     // Check if the benchmark model is available, pull if not
-    const modelsResponse = await axios.get(`${ollamaAPIURL}/api/tags`)
+    const openWebUIService = new (await import('./openwebui_service.js')).OpenWebUIService(this.dockerService)
-    const models = modelsResponse.data.models || []
+    const modelResponse = await openWebUIService.downloadModelSync(AI_BENCHMARK_MODEL)
-    const hasModel = models.some((m: any) => m.name === AI_BENCHMARK_MODEL || m.name.startsWith(AI_BENCHMARK_MODEL.split(':')[0]))
+    if (!modelResponse.success) {
-
+      throw new Error(`Model does not exist and failed to download: ${modelResponse.message}`)
-    if (!hasModel) {
-      this._updateStatus('downloading_ai_model', `Downloading AI benchmark model (${AI_BENCHMARK_MODEL})... This may take a few minutes on first run.`)
-      logger.info(`[BenchmarkService] Model ${AI_BENCHMARK_MODEL} not found, downloading...`)
-
-      try {
-        // Model pull can take several minutes, use longer timeout
-        await axios.post(`${ollamaAPIURL}/api/pull`, { name: AI_BENCHMARK_MODEL }, { timeout: 600000 })
-        logger.info(`[BenchmarkService] Model ${AI_BENCHMARK_MODEL} downloaded successfully`)
-      } catch (pullError) {
-        throw new Error(`Failed to download AI benchmark model (${AI_BENCHMARK_MODEL}): ${pullError.message}`)
-      }
     }
 
     // Run inference benchmark

admin/app/services/docker_service.ts

@@ -13,6 +13,7 @@ export class DockerService {
   private activeInstallations: Set<string> = new Set()
   public static KIWIX_SERVICE_NAME = 'nomad_kiwix_serve'
   public static OLLAMA_SERVICE_NAME = 'nomad_ollama'
+  public static QDRANT_SERVICE_NAME = 'nomad_qdrant'
   public static OPEN_WEBUI_SERVICE_NAME = 'nomad_open_webui'
   public static CYBERCHEF_SERVICE_NAME = 'nomad_cyberchef'
   public static FLATNOTES_SERVICE_NAME = 'nomad_flatnotes'

admin/app/services/openwebui_service.ts

@@ -8,6 +8,9 @@ import path from 'node:path'
 import { PassThrough } from 'node:stream'
 import { DownloadModelJob } from '#jobs/download_model_job'
 import { FALLBACK_RECOMMENDED_OLLAMA_MODELS } from '../../constants/ollama.js'
+import { chromium } from 'playwright'
+import KVStore from '#models/kv_store'
+import { getFile } from '../utils/fs.js'
 
 const NOMAD_MODELS_API_BASE_URL = 'https://api.projectnomad.us/api/v1/ollama/models'
 const MODELS_CACHE_FILE = path.join(process.cwd(), 'storage', 'ollama-models-cache.json')
@@ -15,6 +18,10 @@ const CACHE_MAX_AGE_MS = 24 * 60 * 60 * 1000 // 24 hours
 
 @inject()
 export class OpenWebUIService {
+  public static NOMAD_KNOWLEDGE_BASE_NAME = 'nomad-knowledge-base'
+  public static NOMAD_KNOWLEDGE_BASE_DESCRIP =
+    'Knowledge base managed by Project NOMAD, used to enhance LLM responses with up-to-date information. Do not delete.'
+
   constructor(private dockerService: DockerService) {}
 
   /** We need to call this in the DownloadModelJob, so it can't be private,
@@ -200,6 +207,45 @@ export class OpenWebUIService {
     })
   }
 
+  /**
+   * Synchronous version of model download (waits for completion). Should only be used for
+   * small models or in contexts where a background job is incompatible.
+   * @param model Model name to download
+   * @returns Success status and message
+   */
+  async downloadModelSync(model: string): Promise<{ success: boolean; message: string }> {
+    try {
+      // See if model is already installed
+      const installedModels = await this.getInstalledModels()
+      if (installedModels && installedModels.some((m) => m.name === model)) {
+        logger.info(`[OpenWebUIService] Model "${model}" is already installed.`)
+        return { success: true, message: 'Model is already installed.' }
+      }
+
+      const ollamAPIURL = await this.dockerService.getServiceURL(DockerService.OLLAMA_SERVICE_NAME)
+      if (!ollamAPIURL) {
+        logger.warn('[OpenWebUIService] Ollama service is not running. Cannot download model.')
+        return {
+          success: false,
+          message: 'Ollama is not running. Please start Ollama and try again.',
+        }
+      }
+
+      // 10 minutes timeout for large model downloads
+      await axios.post(`${ollamAPIURL}/api/pull`, { name: model }, { timeout: 600000 })
+
+      logger.info(`[OpenWebUIService] Model "${model}" downloaded via API.`)
+      return { success: true, message: 'Model downloaded successfully.' }
+    } catch (error) {
+      logger.error(
+        `[OpenWebUIService] Failed to download model "${model}": ${
+          error instanceof Error ? error.message : error
+        }`
+      )
+      return { success: false, message: 'Failed to download model.' }
+    }
+  }
+
   async deleteModel(model: string): Promise<{ success: boolean; message: string }> {
     return new Promise((resolve) => {
       try {
@@ -529,6 +575,163 @@ export class OpenWebUIService {
     })
   }
 
+  async getOrCreateKnowledgeBase(): Promise<string | null> {
+    try {
+      // See if we already have the knowledge base ID stored
+      const existing = await KVStore.getValue('open_webui_knowledge_id')
+      if (existing) {
+        return existing as string
+      }
+
+      // Create a new knowledge base via Open WebUI API
+      const tokenData = await this.getOpenWebUIToken()
+      if (!tokenData) {
+        logger.warn(
+          '[OpenWebUIService] Cannot get or create knowledge base because Open WebUI token is unavailable.'
+        )
+        return null
+      }
+
+      const response = await axios.post(
+        `${tokenData.url}/api/v1/knowledge/create`,
+        {
+          name: OpenWebUIService.NOMAD_KNOWLEDGE_BASE_NAME,
+          description: OpenWebUIService.NOMAD_KNOWLEDGE_BASE_DESCRIP,
+        },
+        {
+          headers: {
+            Authorization: `Bearer ${tokenData.token}`,
+          },
+        }
+      )
+
+      if (response.data && response.data.id) {
+        await KVStore.setValue('open_webui_knowledge_id', response.data.id)
+        return response.data.id
+      }
+
+      logger.error(
+        `[OpenWebUIService] Invalid response when creating knowledge base: ${JSON.stringify(
+          response.data
+        )}`
+      )
+      return null
+    } catch (error) {
+      logger.error(
+        `[OpenWebUIService] Failed to get or create knowledge base: ${
+          error instanceof Error ? error.message : error
+        }`
+      )
+      return null
+    }
+  }
+
+  async uploadFileToKnowledgeBase(filepath: string): Promise<boolean> {
+    try {
+      const knowledgeBaseId = await this.getOrCreateKnowledgeBase()
+      if (!knowledgeBaseId) {
+        logger.warn(
+          '[OpenWebUIService] Cannot upload file because knowledge base ID is unavailable and could not be created.'
+        )
+        return false
+      }
+
+      const tokenData = await this.getOpenWebUIToken()
+      if (!tokenData) {
+        logger.warn(
+          '[OpenWebUIService] Cannot upload file because Open WebUI token is unavailable.'
+        )
+        return false
+      }
+
+      const fileStream = await getFile(filepath, 'stream')
+      if (!fileStream) {
+        logger.warn(
+          `[OpenWebUIService] Cannot upload file because it could not be read: ${filepath}`
+        )
+        return false
+      }
+
+      const formData = new FormData()
+      formData.append('file', fileStream)
+
+      const uploadRes = await axios.post(
+        `${tokenData.url}/api/v1/files/`, // Trailing slash seems to be required by OWUI
+        formData,
+        {
+          headers: {
+            'Authorization': `Bearer ${tokenData.token}`,
+            'Content-Type': 'multipart/form-data',
+            'Accept': 'application/json',
+          },
+        }
+      )
+
+      if (!uploadRes.data || !uploadRes.data.id) {
+        logger.error(
+          `[OpenWebUIService] Invalid response when uploading file: ${JSON.stringify(
+            uploadRes.data
+          )}`
+        )
+        return false
+      }
+
+      const fileId = uploadRes.data.id
+
+      // Now associate the uploaded file with the knowledge base
+      const associateRes = await axios.post(
+        `${tokenData.url}/api/v1/knowledge/${knowledgeBaseId}/file/add`,
+        {
+          file_id: fileId,
+        },
+        {
+          headers: {
+            Authorization: `Bearer ${tokenData.token}`,
+          },
+        }
+      )
+
+    } catch (error) {
+      logger.error(
+        `[OpenWebUIService] Failed to upload file to knowledge base: ${
+          error instanceof Error ? error.message : error
+        }`
+      )
+      return false
+    }
+  }
+
+  private async getOpenWebUIToken(): Promise<{ token: string; url: string } | null> {
+    try {
+      const openWebUIURL = await this.dockerService.getServiceURL(
+        DockerService.OPEN_WEBUI_SERVICE_NAME
+      )
+      if (!openWebUIURL) {
+        logger.warn('[OpenWebUIService] Open WebUI service is not running. Cannot retrieve token.')
+        return null
+      }
+      const browser = await chromium.launch({ headless: true })
+      const context = await browser.newContext()
+      const page = await context.newPage()
+
+      await page.goto(openWebUIURL)
+      await page.waitForLoadState('networkidle')
+
+      const cookies = await context.cookies()
+      const tokenCookie = cookies.find((cookie) => cookie.name === 'token')
+      await browser.close()
+
+      return tokenCookie ? { token: tokenCookie.value, url: openWebUIURL } : null
+    } catch (error) {
+      logger.error(
+        `[OpenWebUIService] Failed to retrieve Open WebUI token: ${
+          error instanceof Error ? error.message : error
+        }`
+      )
+      return null
+    }
+  }
+
   private async retrieveAndRefreshModels(
     sort?: 'pulls' | 'name'
   ): Promise<NomadOllamaModel[] | null> {

admin/app/services/rag_service.ts

@@ -0,0 +1,262 @@
+import { Ollama } from 'ollama'
+import { QdrantClient } from '@qdrant/js-client-rest'
+import { DockerService } from './docker_service.js'
+import { inject } from '@adonisjs/core'
+import logger from '@adonisjs/core/services/logger'
+import { chunk } from 'llm-chunk'
+import { OpenWebUIService } from './openwebui_service.js'
+import sharp from 'sharp'
+import { determineFileType, getFile } from '../utils/fs.js'
+import { PDFParse } from 'pdf-parse'
+import { createWorker } from 'tesseract.js'
+import { fromBuffer } from 'pdf2pic'
+
+@inject()
+export class RagService {
+  private qdrant: QdrantClient | null = null
+  private ollama: Ollama | null = null
+  private qdrantInitPromise: Promise<void> | null = null
+  private ollamaInitPromise: Promise<void> | null = null
+  public static CONTENT_COLLECTION_NAME = 'open-webui_knowledge' // This is the collection name OWUI uses for uploaded knowledge
+  public static EMBEDDING_MODEL = 'nomic-embed-text:v1.5'
+  public static EMBEDDING_DIMENSION = 768 // Nomic Embed Text v1.5 dimension is 768
+
+  constructor(
+    private dockerService: DockerService,
+    private openWebUIService: OpenWebUIService
+  ) {}
+
+  private async _initializeQdrantClient() {
+    if (!this.qdrantInitPromise) {
+      this.qdrantInitPromise = (async () => {
+        const qdrantUrl = await this.dockerService.getServiceURL(DockerService.QDRANT_SERVICE_NAME)
+        if (!qdrantUrl) {
+          throw new Error('Qdrant service is not installed or running.')
+        }
+        this.qdrant = new QdrantClient({ url: `http://${qdrantUrl}` })
+      })()
+    }
+    return this.qdrantInitPromise
+  }
+
+  private async _initializeOllamaClient() {
+    if (!this.ollamaInitPromise) {
+      this.ollamaInitPromise = (async () => {
+        const ollamaUrl = await this.dockerService.getServiceURL(DockerService.OLLAMA_SERVICE_NAME)
+        if (!ollamaUrl) {
+          throw new Error('Ollama service is not installed or running.')
+        }
+        this.ollama = new Ollama({ host: `http://${ollamaUrl}` })
+      })()
+    }
+    return this.ollamaInitPromise
+  }
+
+  private async _ensureDependencies() {
+    if (!this.qdrant) {
+      await this._initializeQdrantClient()
+    }
+    if (!this.ollama) {
+      await this._initializeOllamaClient()
+    }
+  }
+
+  private async _ensureCollection(
+    collectionName: string,
+    dimensions: number = RagService.EMBEDDING_DIMENSION
+  ) {
+    try {
+      await this._ensureDependencies()
+      const collections = await this.qdrant!.getCollections()
+      const collectionExists = collections.collections.some((col) => col.name === collectionName)
+
+      if (!collectionExists) {
+        await this.qdrant!.createCollection(collectionName, {
+          vectors: {
+            size: dimensions,
+            distance: 'Cosine',
+          },
+        })
+      }
+    } catch (error) {
+      logger.error('Error ensuring Qdrant collection:', error)
+      throw error
+    }
+  }
+
+  public async embedAndStoreText(
+    text: string,
+    metadata: Record<string, any> = {}
+  ): Promise<{ chunks: number } | null> {
+    try {
+      await this._ensureCollection(
+        RagService.CONTENT_COLLECTION_NAME,
+        RagService.EMBEDDING_DIMENSION
+      )
+      const initModelResponse = await this.openWebUIService.downloadModelSync(
+        RagService.EMBEDDING_MODEL
+      )
+      if (!initModelResponse.success) {
+        throw new Error(
+          `${RagService.EMBEDDING_MODEL} does not exist and could not be downloaded: ${initModelResponse.message}`
+        )
+      }
+
+      const chunks = chunk(text, {
+        // These settings should provide a good balance between context and precision
+        minLength: 512,
+        maxLength: 1024,
+        overlap: 200,
+      })
+
+      if (!chunks || chunks.length === 0) {
+        throw new Error('No text chunks generated for embedding.')
+      }
+
+      const embeddings: number[][] = []
+      for (const chunkText of chunks) {
+        const response = await this.ollama!.embeddings({
+          model: RagService.EMBEDDING_MODEL,
+          prompt: chunkText,
+        })
+
+        embeddings.push(response.embedding)
+      }
+
+      const points = chunks.map((chunkText, index) => ({
+        id: `${Date.now()}_${index}`,
+        vector: embeddings[index],
+        payload: {
+          ...metadata,
+          text: chunkText,
+          chunk_index: index,
+        },
+      }))
+
+      await this.qdrant!.upsert(RagService.CONTENT_COLLECTION_NAME, { points })
+
+      return { chunks: chunks.length }
+    } catch (error) {
+      logger.error('Error embedding text:', error)
+      return null
+    }
+  }
+
+  /**
+   * Preprocess an image to enhance text extraction quality.
+   * Normalizes, grayscales, sharpens, and resizes the image to a manageable size.
+   * @param filebuffer Buffer of the image file
+   * @returns - Processed image buffer
+   */
+  private async preprocessImage(filebuffer: Buffer): Promise<Buffer> {
+    return await sharp(filebuffer)
+      .grayscale()
+      .normalize()
+      .sharpen()
+      .resize({ width: 2000, fit: 'inside' })
+      .toBuffer()
+  }
+
+  /**
+   * If the original PDF has little to no extractable text,
+   * we can use this method to convert each page to an image for OCR processing.
+   * @param filebuffer - Buffer of the PDF file
+   * @returns - Array of image buffers, one per page
+   */
+  private async convertPDFtoImages(filebuffer: Buffer): Promise<Buffer[]> {
+    const converted = await fromBuffer(filebuffer, {
+      quality: 50,
+      density: 200,
+      format: 'png',
+    }).bulk(-1, {
+      responseType: 'buffer',
+    })
+    return converted.filter((res) => res.buffer).map((res) => res.buffer!)
+  }
+
+  /**
+   * Extract text from a PDF file using pdf-parse.
+   * @param filebuffer - Buffer of the PDF file
+   * @returns - Extracted text
+   */
+  private async extractPDFText(filebuffer: Buffer): Promise<string> {
+    const parser = new PDFParse({ data: filebuffer })
+    const data = await parser.getText()
+    await parser.destroy()
+    return data.text
+  }
+
+  /**
+   * Extract text from a plain text file.
+   * @param filebuffer - Buffer of the text file
+   * @returns - Extracted text
+   */
+  private async extractTXTText(filebuffer: Buffer): Promise<string> {
+    return filebuffer.toString('utf-8')
+  }
+
+  /**
+   * Extract text from an image file using Tesseract.js OCR.
+   * @param filebuffer - Buffer of the image file
+   * @returns - Extracted text
+   */
+  private async extractImageText(filebuffer: Buffer): Promise<string> {
+    const worker = await createWorker('eng')
+    const result = await worker.recognize(filebuffer)
+    await worker.terminate()
+    return result.data.text
+  }
+
+  /**
+   * Main pipeline to process and embed an uploaded file into the RAG knowledge base.
+   * This includes text extraction, chunking, embedding, and storing in Qdrant.
+   */
+  public async processAndEmbedFile(
+    filepath: string
+  ): Promise<{ success: boolean; message: string }> {
+    try {
+      const fileType = determineFileType(filepath)
+      if (fileType === 'unknown') {
+        return { success: false, message: 'Unsupported file type.' }
+      }
+
+      const origFileBuffer = await getFile(filepath, 'buffer')
+      if (!origFileBuffer) {
+        return { success: false, message: 'Failed to read the uploaded file.' }
+      }
+
+      let extractedText = ''
+
+      if (fileType === 'image') {
+        const preprocessedBuffer = await this.preprocessImage(origFileBuffer)
+        extractedText = await this.extractImageText(preprocessedBuffer)
+      } else if (fileType === 'pdf') {
+        extractedText = await this.extractPDFText(origFileBuffer)
+        // Check if there was no extracted text or it was very minimal
+        if (!extractedText || extractedText.trim().length < 100) {
+          // Convert PDF pages to images for OCR
+          const imageBuffers = await this.convertPDFtoImages(origFileBuffer)
+          for (const imgBuffer of imageBuffers) {
+            const preprocessedImg = await this.preprocessImage(imgBuffer)
+            const pageText = await this.extractImageText(preprocessedImg)
+            extractedText += pageText + '\n'
+          }
+        }
+      } else {
+        extractedText = await this.extractTXTText(origFileBuffer)
+      }
+
+      if (!extractedText || extractedText.trim().length === 0) {
+        return { success: false, message: 'No text could be extracted from the file.' }
+      }
+
+      const embedResult = await this.embedAndStoreText(extractedText, {})
+
+
+      return { success: true, message: 'File processed and embedded successfully.' }
+    } catch (error) {
+      logger.error('Error processing and embedding file:', error)
+      return { success: false, message: 'Error processing and embedding file.' }
+    }
+  }
+}

admin/app/utils/fs.ts

@@ -1,5 +1,5 @@
 import { mkdir, readdir, readFile, stat, unlink } from 'fs/promises'
-import { join } from 'path'
+import path, { join } from 'path'
 import { FileEntry } from '../../types/files.js'
 import { createReadStream } from 'fs'
 import { LSBlockDevice, NomadDiskInfoRaw } from '../../types/system.js'
@@ -151,3 +151,16 @@ export function matchesDevice(fsPath: string, deviceName: string): boolean {
 
   return false
 }
+
+export function determineFileType(filename: string): 'image' | 'pdf' | 'text' | 'unknown' {
+  const ext = path.extname(filename).toLowerCase()
+  if (['.jpg', '.jpeg', '.png', '.gif', '.bmp', '.tiff', '.webp'].includes(ext)) {
+    return 'image'
+  } else if (ext === '.pdf') {
+    return 'pdf'
+  } else if (['.txt', '.md', '.docx', '.rtf'].includes(ext)) {
+    return 'text'
+  } else {
+    return 'unknown'
+  }
+}

admin/database/migrations/1769400000002_create_kv_store_table.ts

@@ -0,0 +1,19 @@
+import { BaseSchema } from '@adonisjs/lucid/schema'
+
+export default class extends BaseSchema {
+  protected tableName = 'kv_store'
+
+  async up() {
+    this.schema.createTable(this.tableName, (table) => {
+      table.increments('id')
+      table.string('key').unique().notNullable()
+      table.text('value').nullable()
+      table.timestamp('created_at')
+      table.timestamp('updated_at')
+    })
+  }
+
+  async down() {
+    this.schema.dropTable(this.tableName)
+  }
+}

admin/database/seeders/service_seeder.ts

@@ -3,17 +3,25 @@ import { DockerService } from '#services/docker_service'
 import { BaseSeeder } from '@adonisjs/lucid/seeders'
 import { ModelAttributes } from '@adonisjs/lucid/types/model'
 import env from '#start/env'
+import { RagService } from '#services/rag_service'
 
 export default class ServiceSeeder extends BaseSeeder {
   // Use environment variable with fallback to production default
-  private static NOMAD_STORAGE_ABS_PATH = env.get('NOMAD_STORAGE_PATH', '/opt/project-nomad/storage')
+  private static NOMAD_STORAGE_ABS_PATH = env.get(
-  private static DEFAULT_SERVICES: Omit<ModelAttributes<Service>, 'created_at' | 'updated_at' | 'metadata' | 'id'>[] = [
+    'NOMAD_STORAGE_PATH',
+    '/opt/project-nomad/storage'
+  )
+  private static DEFAULT_SERVICES: Omit<
+    ModelAttributes<Service>,
+    'created_at' | 'updated_at' | 'metadata' | 'id'
+  >[] = [
     {
       service_name: DockerService.KIWIX_SERVICE_NAME,
       friendly_name: 'Information Library',
       powered_by: 'Kiwix',
       display_order: 1,
-      description: 'Offline access to Wikipedia, medical references, how-to guides, and encyclopedias',
+      description:
+        'Offline access to Wikipedia, medical references, how-to guides, and encyclopedias',
       icon: 'IconBooks',
       container_image: 'ghcr.io/kiwix/kiwix-serve:3.8.1',
       container_command: '*.zim --address=all',
@@ -21,9 +29,9 @@ export default class ServiceSeeder extends BaseSeeder {
         HostConfig: {
           RestartPolicy: { Name: 'unless-stopped' },
           Binds: [`${ServiceSeeder.NOMAD_STORAGE_ABS_PATH}/zim:/data`],
-          PortBindings: { '8080/tcp': [{ HostPort: '8090' }] }
+          PortBindings: { '8080/tcp': [{ HostPort: '8090' }] },
         },
-        ExposedPorts: { '8080/tcp': {} }
+        ExposedPorts: { '8080/tcp': {} },
       }),
       ui_location: '8090',
       installed: false,
@@ -31,6 +39,29 @@ export default class ServiceSeeder extends BaseSeeder {
       is_dependency_service: false,
       depends_on: null,
     },
+    {
+      service_name: DockerService.QDRANT_SERVICE_NAME,
+      friendly_name: 'Qdrant Vector Database',
+      powered_by: null,
+      display_order: 100, // Dependency service, not shown directly
+      description: 'Vector database for storing and searching embeddings',
+      icon: 'IconRobot',
+      container_image: 'qdrant/qdrant:latest',
+      container_command: null,
+      container_config: JSON.stringify({
+        HostConfig: {
+          RestartPolicy: { Name: 'unless-stopped' },
+          Binds: [`${ServiceSeeder.NOMAD_STORAGE_ABS_PATH}/qdrant:/qdrant/storage`],
+          PortBindings: { '6333/tcp': [{ HostPort: '6333' }], '6334/tcp': [{ HostPort: '6334' }] },
+        },
+        ExposedPorts: { '6333/tcp': {}, '6334/tcp': {} },
+      }),
+      ui_location: '6333',
+      installed: false,
+      installation_status: 'idle',
+      is_dependency_service: true,
+      depends_on: null,
+    },
     {
       service_name: DockerService.OLLAMA_SERVICE_NAME,
       friendly_name: 'Ollama',
@@ -44,15 +75,15 @@ export default class ServiceSeeder extends BaseSeeder {
         HostConfig: {
           RestartPolicy: { Name: 'unless-stopped' },
           Binds: [`${ServiceSeeder.NOMAD_STORAGE_ABS_PATH}/ollama:/root/.ollama`],
-          PortBindings: { '11434/tcp': [{ HostPort: '11434' }] }
+          PortBindings: { '11434/tcp': [{ HostPort: '11434' }] },
         },
-        ExposedPorts: { '11434/tcp': {} }
+        ExposedPorts: { '11434/tcp': {} },
       }),
       ui_location: null,
       installed: false,
       installation_status: 'idle',
       is_dependency_service: true,
-      depends_on: null,
+      depends_on: DockerService.QDRANT_SERVICE_NAME,
     },
     {
       service_name: DockerService.OPEN_WEBUI_SERVICE_NAME,
@@ -68,9 +99,17 @@ export default class ServiceSeeder extends BaseSeeder {
           RestartPolicy: { Name: 'unless-stopped' },
           NetworkMode: 'host',
           Binds: [`${ServiceSeeder.NOMAD_STORAGE_ABS_PATH}/open-webui:/app/backend/data`],
-          PortBindings: { '8080/tcp': [{ HostPort: '3000' }] }
+          PortBindings: { '8080/tcp': [{ HostPort: '3000' }] },
         },
-        Env: ['WEBUI_AUTH=False', 'PORT=3000', 'OLLAMA_BASE_URL=http://127.0.0.1:11434']
+        Env: [
+          'WEBUI_AUTH=False',
+          'PORT=3000',
+          'OLLAMA_BASE_URL=http://127.0.0.1:11434',
+          'VECTOR_DB=qdrant',
+          'QDRANT_URI=http://127.0.0.1:6333',
+          'RAG_EMBEDDING_ENGINE=ollama',
+          `RAG_EMBEDDING_MODEL=${RagService.EMBEDDING_MODEL}`,
+        ],
       }),
       ui_location: '3000',
       installed: false,
@@ -90,9 +129,9 @@ export default class ServiceSeeder extends BaseSeeder {
       container_config: JSON.stringify({
         HostConfig: {
           RestartPolicy: { Name: 'unless-stopped' },
-          PortBindings: { '80/tcp': [{ HostPort: '8100' }] }
+          PortBindings: { '80/tcp': [{ HostPort: '8100' }] },
         },
-        ExposedPorts: { '80/tcp': {} }
+        ExposedPorts: { '80/tcp': {} },
       }),
       ui_location: '8100',
       installed: false,
@@ -113,10 +152,10 @@ export default class ServiceSeeder extends BaseSeeder {
         HostConfig: {
           RestartPolicy: { Name: 'unless-stopped' },
           PortBindings: { '8080/tcp': [{ HostPort: '8200' }] },
-          Binds: [`${ServiceSeeder.NOMAD_STORAGE_ABS_PATH}/flatnotes:/data`]
+          Binds: [`${ServiceSeeder.NOMAD_STORAGE_ABS_PATH}/flatnotes:/data`],
         },
         ExposedPorts: { '8080/tcp': {} },
-        Env: ['FLATNOTES_AUTH_TYPE=none']
+        Env: ['FLATNOTES_AUTH_TYPE=none'],
       }),
       ui_location: '8200',
       installed: false,
@@ -137,7 +176,7 @@ export default class ServiceSeeder extends BaseSeeder {
         HostConfig: {
           RestartPolicy: { Name: 'unless-stopped' },
           PortBindings: { '8080/tcp': [{ HostPort: '8300' }] },
-          Binds: [`${ServiceSeeder.NOMAD_STORAGE_ABS_PATH}/kolibri:/root/.kolibri`]
+          Binds: [`${ServiceSeeder.NOMAD_STORAGE_ABS_PATH}/kolibri:/root/.kolibri`],
         },
         ExposedPorts: { '8080/tcp': {} },
       }),
@@ -146,15 +185,17 @@ export default class ServiceSeeder extends BaseSeeder {
       installation_status: 'idle',
       is_dependency_service: false,
       depends_on: null,
-    }
+    },
   ]
 
   async run() {
     const existingServices = await Service.query().select('service_name')
-    const existingServiceNames = new Set(existingServices.map(service => service.service_name))
+    const existingServiceNames = new Set(existingServices.map((service) => service.service_name))
 
-    const newServices = ServiceSeeder.DEFAULT_SERVICES.filter(service => !existingServiceNames.has(service.service_name))
+    const newServices = ServiceSeeder.DEFAULT_SERVICES.filter(
+      (service) => !existingServiceNames.has(service.service_name)
+    )
 
     await Service.createMany([...newServices])
   }
-}
+}

admin/inertia/components/file-uploader/index.css

@@ -0,0 +1,3 @@
+.uppy-size--md .uppy-Dashboard-AddFiles-title {
+  font-size: 1.15rem !important;
+}

admin/inertia/components/file-uploader/index.tsx

@@ -0,0 +1,84 @@
+import React, { useState } from 'react'
+import Uppy from '@uppy/core'
+import '@uppy/core/css/style.min.css'
+import '@uppy/dashboard/css/style.min.css'
+import { useUppyEvent } from '@uppy/react'
+import Dashboard from '@uppy/react/dashboard'
+import classNames from 'classnames'
+import './index.css' // Custom styles for the uploader
+
+interface FileUploaderProps {
+  minFiles?: number // minimum number of files required
+  maxFiles?: number
+  maxFileSize?: number // in bytes, e.g., 10485760 for 10MB
+  fileTypes?: string[] // e.g., ['image/*', 'application/pdf']
+  disabled?: boolean
+  onUpload: (files: FileList) => void
+  className?: string
+}
+
+/**
+ * A drag-and-drop (or click) file upload area with customizations for
+ * multiple and maximum numbers of files.
+ */
+const FileUploader: React.FC<FileUploaderProps> = ({
+  minFiles = 0,
+  maxFiles = 1,
+  maxFileSize = 10485760, // default to 10MB
+  fileTypes,
+  disabled = false,
+  onUpload,
+  className,
+}) => {
+  const [uppy] = useState(() => {
+    const uppy = new Uppy({
+      debug: true,
+      restrictions: {
+        maxFileSize: maxFileSize,
+        minNumberOfFiles: minFiles,
+        maxNumberOfFiles: maxFiles,
+        allowedFileTypes: fileTypes || undefined,
+      },
+    })
+    return uppy
+  })
+
+  useUppyEvent(uppy, 'state-update', (_, newState) => {
+    const stateFiles = Object.values(newState.files)
+
+    const dataTransfer = new DataTransfer()
+    stateFiles.forEach((file) => {
+      if (file.data) {
+        if (file.data instanceof File) {
+          dataTransfer.items.add(file.data)
+        } else if (file.data instanceof Blob) {
+          const newFile = new File(
+            [file.data],
+            file.name || `${crypto.randomUUID()}.${file.extension}`,
+            {
+              type: file.type,
+              lastModified: new Date().getTime(),
+            }
+          )
+          dataTransfer.items.add(newFile)
+        }
+      }
+    })
+
+    const fileList = dataTransfer.files
+    onUpload(fileList) // Always send new file list even if empty
+  })
+
+  return (
+    <Dashboard
+      uppy={uppy}
+      width={'100%'}
+      height={'250px'}
+      hideUploadButton
+      disabled={disabled}
+      className={classNames(className)}
+    />
+  )
+}
+
+export default FileUploader

admin/inertia/pages/home.tsx

@@ -1,5 +1,6 @@
 import {
   IconBolt,
+  IconBrain,
   IconHelp,
   IconMapRoute,
   IconPlus,
@@ -80,6 +81,17 @@ interface DashboardItem {
   poweredBy: string | null
 }
 
+const KNOWLEDGE_BASE_ITEM: DashboardItem = {
+  label: 'Knowledge Base',
+  to: '/knowledge-base',
+  target: '',
+  description: 'Upload documents to your personal knowledge base for AI access',
+  icon: <IconBrain size={48} />,
+  installed: true,
+  displayOrder: 5,
+  poweredBy: null,
+}
+
 export default function Home(props: {
   system: {
     services: ServiceSlim[]
@@ -114,6 +126,9 @@ export default function Home(props: {
 
   // Add system items
   items.push(...SYSTEM_ITEMS)
+  if (props.system.services.find((s) => s.service_name === 'nomad_open_webui' && s.installed)) {
+    items.push(KNOWLEDGE_BASE_ITEM)
+  }
 
   // Sort all items by display order
   items.sort((a, b) => a.displayOrder - b.displayOrder)
@@ -130,9 +145,7 @@ export default function Home(props: {
             >
               <div className="flex items-center justify-center mb-2">{item.icon}</div>
               <h3 className="font-bold text-2xl">{item.label}</h3>
-              {item.poweredBy && (
+              {item.poweredBy && <p className="text-sm opacity-80">Powered by {item.poweredBy}</p>}
-                <p className="text-sm opacity-80">Powered by {item.poweredBy}</p>
-              )}
               <p className="xl:text-lg mt-2">{item.description}</p>
             </div>
           </a>

admin/inertia/pages/knowledge-base.tsx

@@ -0,0 +1,97 @@
+import { Head } from '@inertiajs/react'
+import { useState } from 'react'
+import FileUploader from '~/components/file-uploader'
+import StyledButton from '~/components/StyledButton'
+import AppLayout from '~/layouts/AppLayout'
+
+export default function KnowledgeBase() {
+  const [loading, setLoading] = useState(false)
+  const [files, setFiles] = useState<File[]>([])
+
+  return (
+    <AppLayout>
+      <Head title="Knowledge Base" />
+      <main className="px-6 lg:px-12 py-6 lg:py-8">
+        <div className="bg-white rounded-lg border shadow-md overflow-hidden">
+          <div className="p-6">
+            <FileUploader
+              minFiles={1}
+              maxFiles={1}
+              onUpload={(files) => {
+                setLoading(true)
+                setFiles(Array.from(files))
+                setLoading(false)
+              }}
+            />
+            <div className="flex justify-center gap-4 my-6">
+              <StyledButton
+                variant="primary"
+                size="lg"
+                icon="ArrowUpCircleIcon"
+                onClick={() => {}}
+                disabled={files.length === 0 || loading}
+                loading={loading}
+              >
+                Upload
+              </StyledButton>
+            </div>
+          </div>
+          <div className="border-t bg-white p-6">
+            <h3 className="text-lg font-semibold text-desert-green mb-4">
+              Why upload documents to your Knowledge Base?
+            </h3>
+            <div className="space-y-3">
+              <div className="flex items-start gap-3">
+                <div className="flex-shrink-0 w-6 h-6 rounded-full bg-desert-green text-white flex items-center justify-center text-sm font-bold">
+                  1
+                </div>
+                <div>
+                  <p className="font-medium text-desert-stone-dark">
+                    AI Assistant Knowledge Base Integration
+                  </p>
+                  <p className="text-sm text-desert-stone">
+                    When you upload documents to your Knowledge Base, NOMAD processes and embeds the
+                    content, making it directly accessible to the AI Assistant. This allows the AI
+                    Assistant to reference your specific documents during conversations, providing
+                    more accurate and personalized responses based on your uploaded data.
+                  </p>
+                </div>
+              </div>
+              <div className="flex items-start gap-3">
+                <div className="flex-shrink-0 w-6 h-6 rounded-full bg-desert-green text-white flex items-center justify-center text-sm font-bold">
+                  2
+                </div>
+                <div>
+                  <p className="font-medium text-desert-stone-dark">
+                    Enhanced Document Processing with OCR
+                  </p>
+                  <p className="text-sm text-desert-stone">
+                    NOMAD includes built-in Optical Character Recognition (OCR) capabilities,
+                    allowing it to extract text from image-based documents such as scanned PDFs or
+                    photos. This means that even if your documents are not in a standard text
+                    format, NOMAD can still process and embed their content for AI access.
+                  </p>
+                </div>
+              </div>
+              <div className="flex items-start gap-3">
+                <div className="flex-shrink-0 w-6 h-6 rounded-full bg-desert-green text-white flex items-center justify-center text-sm font-bold">
+                  3
+                </div>
+                <div>
+                  <p className="font-medium text-desert-stone-dark">
+                    Information Library Integration
+                  </p>
+                  <p className="text-sm text-desert-stone">
+                    NOMAD will automatically discover and extract any content you save to your
+                    Information Library (if installed), making it instantly available to the AI
+                    Assistant without any extra steps.
+                  </p>
+                </div>
+              </div>
+            </div>
+          </div>
+        </div>
+      </main>
+    </AppLayout>
+  )
+}

admin/package.json

@@ -76,11 +76,15 @@
     "@inertiajs/react": "^2.0.13",
     "@markdoc/markdoc": "^0.5.2",
     "@protomaps/basemaps": "^5.7.0",
+    "@qdrant/js-client-rest": "^1.16.2",
     "@tabler/icons-react": "^3.34.0",
     "@tailwindcss/vite": "^4.1.10",
     "@tanstack/react-query": "^5.81.5",
     "@tanstack/react-query-devtools": "^5.83.0",
     "@tanstack/react-virtual": "^3.13.12",
+    "@uppy/core": "^5.2.0",
+    "@uppy/dashboard": "^5.1.0",
+    "@uppy/react": "^5.1.1",
     "@vinejs/vine": "^3.0.1",
     "@vitejs/plugin-react": "^4.6.0",
     "autoprefixer": "^10.4.21",
@@ -90,10 +94,15 @@
     "dockerode": "^4.0.7",
     "edge.js": "^6.2.1",
     "fast-xml-parser": "^5.2.5",
+    "llm-chunk": "^0.0.1",
     "luxon": "^3.6.1",
     "maplibre-gl": "^4.7.1",
     "mysql2": "^3.14.1",
+    "ollama": "^0.6.3",
+    "pdf-parse": "^2.4.5",
+    "pdf2pic": "^3.2.0",
     "pino-pretty": "^13.0.0",
+    "playwright": "^1.58.0",
     "pmtiles": "^4.3.0",
     "postcss": "^8.5.6",
     "react": "^19.1.0",
@@ -101,9 +110,11 @@
     "react-dom": "^19.1.0",
     "react-map-gl": "^8.1.0",
     "reflect-metadata": "^0.2.2",
+    "sharp": "^0.34.5",
     "systeminformation": "^5.27.14",
     "tailwindcss": "^4.1.10",
     "tar": "^7.5.6",
+    "tesseract.js": "^7.0.0",
     "url-join": "^5.0.0",
     "usehooks-ts": "^3.1.1",
     "yaml": "^2.8.0"

admin/start/routes.ts

@@ -24,6 +24,7 @@ transmit.registerRoutes()
 router.get('/', [HomeController, 'index'])
 router.get('/home', [HomeController, 'home'])
 router.on('/about').renderInertia('about')
+router.on('/knowledge-base').renderInertia('knowledge-base')
 router.get('/maps', [MapsController, 'index'])
 
 router.get('/easy-setup', [EasySetupController, 'index'])

admin/types/kv_store.ts

@@ -0,0 +1,3 @@
+
+export type KVStoreKey = 'open_webui_knowledge_id'
+export type KVStoreValue = string | null

admin/types/ollama.ts

@@ -27,3 +27,13 @@ export type OllamaModelListing = {
   size: string
   modified: string
 }
+
+
+export type OpenWebUIKnowledgeFileMetadata = {
+  source: string
+  name: string
+  created_by: string
+  file_id: string
+  start_index: number
+  hash: string
+}