About 5 min read

Knowledge Base

Karma One's Knowledge Base lets you upload your own documents and files so the AI can reference them during conversations. Think of it as giving your AI assistant a personal library.

Overview

The Knowledge Base system allows you to:

  • Upload documents in a wide range of formats
  • Query your documents using natural language
  • Search across multiple knowledge bases simultaneously
  • Keep your data organized, updated, and private

Whether you are building a research library, a company wiki, or a personal reference collection, the Knowledge Base makes your information instantly accessible through conversation.

Three Search Engines

Under the hood, Karma One supports three different search engines. Each has its strengths, and you can mix them freely.

| Engine | ID Prefix | Best For | Technology | |--------|-----------|----------|------------| | Gemini File Search | fileSearchStores/ | General-purpose document search, large file collections | Google Gemini embedding and retrieval | | Bubble RAG | bubble-rag: | High-precision retrieval, technical documentation | Retrieval-Augmented Generation pipeline | | Bubble File Search (BFS) | bfs: | Rich media files, per-document deep queries | Semantic tree + vector search |

You do not need to choose an engine manually -- Karma One routes queries to the appropriate engine based on the knowledge base you are using.

Supported File Formats

| Category | Formats | |----------|---------| | Documents | PDF, DOC, DOCX, PPT, PPTX, XLS, XLSX | | Text | TXT, MD (Markdown), CSV, JSON, JSONL | | Images | PNG, JPG, JPEG, GIF, BMP, WEBP, SVG, TIFF | | Audio | MP3, WAV, FLAC, AAC, OGG, WMA, M4A | | Video | MP4, AVI, MOV, MKV, WMV, FLV, WEBM | | Code | Any plain-text source file (.py, .js, .ts, .go, etc.) |

Note: Images, audio, and video files are processed through AI-powered analysis to extract searchable content (e.g., text in images via OCR, transcription for audio).

Creating a Knowledge Base

  1. Open the Knowledge Base panel from the sidebar or the Karma One configuration screen.
  2. Tap Create New Knowledge Base.
  3. Give it a descriptive name (e.g., "Product Specs Q1 2026", "Research Papers", "Company Handbook").
  4. The knowledge base is created and ready for uploads.

Uploading Documents

From the UI

  1. Open the knowledge base you want to add files to.
  2. Tap the Upload button.
  3. Select one or more files from your device.
  4. Wait for processing to complete (a progress indicator shows the status).

Processing Pipeline

After upload, each file goes through an automated pipeline:

Upload -> Parse -> Chunk -> Summarize -> Vectorize -> Index -> Ready

| Stage | What Happens | |-------|-------------| | Parse | Extracts raw text from the file format | | Chunk | Splits the content into semantically meaningful segments | | Summarize | Generates a summary for each chunk | | Vectorize | Creates embedding vectors for semantic search | | Index | Builds a semantic tree for fast retrieval |

Processing time varies by file size and type. A typical PDF takes 10-30 seconds; large files or media may take longer.

Using Knowledge Bases in Conversations

Once your documents are uploaded and indexed, you can query them naturally:

"What does our company handbook say about remote work policies?"
"Find the section about error handling in the API documentation."
"Summarize the key findings from the Q4 market research report."

Enabling a Knowledge Base

To use a specific knowledge base in a conversation:

  1. Open the conversation settings or the Karma One configuration panel.
  2. Select one or more knowledge bases from the list.
  3. The AI will now search those knowledge bases when relevant.

Alternatively, you can tell the AI directly:

"Search my 'Product Specs' knowledge base for battery life information."

Multi-KB Simultaneous Query

Karma One can search multiple knowledge bases at the same time and merge the results. This is powerful when your information is spread across different collections.

"Compare what the engineering docs and the marketing brief say about our new feature."

The AI queries both knowledge bases in parallel, retrieves the most relevant chunks, and synthesizes a unified answer with citations indicating which knowledge base each piece of information came from.

Management

Renaming a Knowledge Base

  1. Open the knowledge base.
  2. Tap the settings or edit icon.
  3. Enter a new name and confirm.

Deleting a Knowledge Base

  1. Open the knowledge base.
  2. Tap Delete in the settings menu.
  3. Confirm the deletion.

Warning: Deleting a knowledge base permanently removes all uploaded files and their indexes. This action cannot be undone.

Updating Documents

To update a document:

  1. Delete the outdated version from the knowledge base.
  2. Upload the new version.
  3. Wait for reprocessing to complete.

The knowledge base will reflect the updated content once indexing is finished.

Removing Individual Files

You can remove specific files from a knowledge base without deleting the entire collection. Open the knowledge base, find the file, and tap the delete icon.

Best Practices

  • Organize by topic. Create separate knowledge bases for different projects or domains rather than dumping everything into one.
  • Use descriptive names. "Q1 2026 Sales Data" is better than "Data Files" when you have multiple knowledge bases.
  • Keep documents current. Outdated information can lead to incorrect AI responses. Remove or replace files when they become stale.
  • Start small, expand gradually. Upload a few key documents first and test the retrieval quality before adding hundreds of files.
  • Leverage multi-KB queries. If you are working across domains (e.g., technical specs + business requirements), select both knowledge bases for richer context.
  • Check citations. When the AI references a knowledge base, it indicates the source. Verify critical information against the original document.

Limitations

| Limitation | Details | |------------|---------| | File size | Individual files are subject to upload size limits based on your plan | | Processing time | Large files and media (audio/video) take longer to process | | Retrieval accuracy | Works best with well-structured documents; scanned images with poor OCR quality may yield lower accuracy | | Real-time updates | Knowledge bases are not live-synced; you must re-upload files when they change |