Comprehensive system for working with vector databases. Upload documents, process content with AI, vectorize and search semantically. Full integration with self-learning system.
Why manual knowledge base population is painful
Documents in PDF, Word, Excel, presentations... Each format needs its own parser. Collecting everything together takes hours of manual work.
Is splitting text into chunks simple? No. Breaking in the middle of a thought kills semantics. Chunks too large mean poor search.
Without quality embeddings and proper indexes, searching the knowledge base becomes a lottery.
Some documents in Google Drive, some in Dropbox, something in OneDrive. Gathering everything in one place is a quest.
Updated a document — how to update the database entry? Delete old, add new, edit existing?
How many records in the database? Which tables are updated most actively? How is self-learning working? Complete unknown.
Everything you need to manage AI agent knowledge
PDF, DOCX, XLSX, CSV, TXT, PPTX — all popular document formats. Upload files directly or via URL from cloud storage.
Intelligent fragmentation with Gemini AI. Automatic metadata generation: categories, keywords, topics.
Embeddings via Google Gemini Embeddings. HNSW indexes in Supabase for lightning-fast semantic search.
Structured tables in PostgreSQL + vectorization in Supabase. Maximum flexibility for any tasks.
API key authorization, rate limiting (30 req/min), table and mode validation. Protection from unauthorized access.
Detailed self-learning statistics: cycles, dialogues, approved/rejected updates, activity by tables.
Full control over data in vector databases
Add new records to the end of the table without deleting existing ones. Ideal for gradually building up the knowledge base.
Safe additionComplete table cleanup and replacement of all data with new content. For full knowledge base rewrite.
Full rewriteUpdate a specific record by ID. For pinpoint changes to individual records without affecting others.
Requires editIdDelete record by ID with cascading deletion from related tables (document_rows, document_metadata).
Cascading deleteSupport for all popular document formats
Extract PDF Text node
Direct text extraction
ZIP → XML → Parse
Dual storage mode
Table + Vector
Extract slides text
Automatic URL recognition and conversion
docs.google.com
→ DOCX exportspreadsheets
→ XLSX exportpresentation
→ PPTX exportdrive.google.com
→ Direct downloaddropbox.com
→ dl=1 parameteronedrive.live.com
→ Direct supportIntelligent fragmentation and metadata generation
Supabase + HNSW indexes for semantic search
Product knowledge base — prices, features, algorithms
Interaction policies — channels, rules, limitations
Sales strategies — argumentation, objections, closing
Contact capture methods — progressive disclosure
Dialogue scenarios — use cases, escalation, branching
Full synchronization with AI agent self-learning system
The self-learning system analyzes dialogues, generates prompt improvement suggestions, validates them, and applies them to vector databases through Vector Base Writer. AI Core Agent immediately starts using updated knowledge. The /get-learning-stats endpoint shows detailed statistics of the entire process.
Why Vector Base Writer is the right choice
One API call to upload a document of any format. The system will detect the type and process it automatically.
AI preserves semantic integrity of text. Related thoughts are not broken in the middle of a sentence.
768-dimensional embeddings and HNSW indexes provide lightning-fast semantic search.
4 operation modes — append, replace, edit, delete. Full control over data.
Stats endpoint shows everything: learning cycles, updates, activity by tables.
API keys, rate limiting, validation. Your data is securely protected.
Key system parameters