From document chaos to instant answers in minutes.
Here's exactly what happens when you connect Knowra to your documents and your AI.
Why connecting Claude directly to Google Drive isn't enough.
Every major AI platform now offers a way to connect to Google Drive or OneDrive. The promise is compelling — ask your AI about your documents and get answers. The reality is more limited than it sounds.
Native cloud drive connections work by finding files and reading their content. For a small number of short documents that's fine. But for a team with hundreds or thousands of files, the AI either can't find the right document, reads entire files burning through your context window, or returns answers based on whatever it happens to retrieve — not necessarily the most relevant information.
Knowra takes a different approach. Instead of finding and reading files at query time, Knowra pre-indexes your documents into a semantic knowledge base. When you ask a question, Knowra retrieves only the most relevant passages and passes that focused context to your AI. The result is faster, more accurate answers that scale to thousands of documents.
What happens when you connect a folder.
OAuth connection
You authorise Knowra to read your selected folder via Google, Microsoft, or Dropbox OAuth. Your credentials are encrypted and stored securely. Knowra never writes to your storage — read access only.
Document extraction
Knowra downloads each file and extracts its text content. Supported: PDF, Word documents, PowerPoint, plain text, Google Docs, and Google Sheets.
Structure-aware chunking, then embedding
Each document is split at natural boundaries — paragraphs, then sentences — so no chunk cuts a thought in half, into pieces of roughly 200–400 words with overlap between neighbouring chunks to preserve context across a split. Each chunk is converted into a 1536-dimension vector embedding using Google's Gemini embedding model (gemini-embedding-001).
Vector index storage
Chunks and their embeddings are stored in a PostgreSQL database with pgvector, indexed with HNSW for fast approximate nearest-neighbour search. On Business plans, your index re-syncs automatically every day. On all other plans, sync is triggered manually from your dashboard whenever you need it.
What happens when you ask a question.
Your question is embedded
When your AI calls Knowra's MCP endpoint with your question, Knowra converts the question into the same vector format as your document chunks.
Semantic similarity search
Knowra finds the chunks whose meaning is most similar to your question, ranked by cosine similarity between vector embeddings — not just matching keywords, but understanding intent. A question about 'cold storage requirements' finds chunks about '-20°C conditions' even without those exact words.
Grounded answer generation
The most relevant chunks are returned to your AI client as context. Your AI generates its answer from that specific, cited content rather than from general training data.
What is MCP and why does it matter?
MCP — Model Context Protocol — is an open standard that lets AI assistants connect to external tools and data sources. Supported by Anthropic, OpenAI, and Google, MCP allows Claude, ChatGPT, and Gemini to call external services to retrieve information during a conversation.
Knowra exposes your knowledge base as an MCP server. When you connect Knowra to your AI client, your AI automatically searches your documents whenever a question might be answered by your internal knowledge — transparently, on every relevant query.
Your documents stay yours.
OAuth read-only access
Knowra never writes to your cloud storage. Revoke access anytime from your provider's security settings.
Encrypted credentials
OAuth tokens are encrypted at rest. Raw credentials are never stored.
Workspace isolation
Each Knowra workspace is completely isolated. No data is shared between organisations.
No training on your data
Your documents are never used to train any AI model.