Knowledge Base Overview

Upload, store, and ingest enterprise documents for real-time RAG context in Voice AI and messaging

Knowledge Base API

The Knowledge Base API enables you to upload, manage, and retrieve proprietary company documentation. Uploaded files (PDF, DOCX, TXT) are automatically parsed, chunked, converted into vector embeddings, and stored in Sayvy AI’s vector database. During live Voice AI phone conversations and automated WhatsApp messaging, agents query this knowledge base via semantic search to deliver accurate, grounded answers in real time.

+-------------------------------------------------------------+
|               Knowledge Base Ingestion Pipeline             |
|                                                             |
|   +-----------------------+     +-----------------------+   |
|   | Corporate Documents   |     | Product & Policy Docs |   |
|   |  (PDF / DOCX / TXT)   |     |  (Pricing / FAQs / KB)|   |
|   +-----------+-----------+     +-----------+-----------+   |
|               \                             /               |
|                v                           v                |
|              +-------------------------------+              |
|              |     Document Parser & Chunker |              |
|              |    (Extract text & paragraphs)|              |
|              +---------------+---------------+              |
|                              |                              |
|                              v                              |
|              +-------------------------------+              |
|              |    Vector Embedding Engine    |              |
|              |  (High-dimensional embeddings)|              |
|              +---------------+---------------+              |
|                              |                              |
|              +---------------+---------------+              |
|              |                               |              |
|              v                               v              |
|     +-----------------+             +-----------------+     |
|     |  Cloud Storage  |             | Vector Database |     |
|     | (R2 / S3 Store) |             | (Semantic RAG)  |     |
|     +-----------------+             +-----------------+     |
+-------------------------------------------------------------+

---

Key Capabilities

Multi-Format Ingestion

Upload standard documentation formats including Portable Document Format (PDF), Microsoft Word (DOCX), and plain text (TXT).

Automated Vectorization

Background workers chunk document content and generate high-precision vector embeddings for instant semantic search.

Secure Cloud Storage

Source document files are encrypted and securely stored in S3/R2 object storage with streaming download support.

Voice AI RAG Integration

Conversational Voice agents reference document vectors mid-call to answer caller inquiries without hallucinating.

---

Supported File Formats

Format	Extension	Max File Size	Processing Pipeline
Plain Text	.txt	10 MB	Direct UTF-8 text extraction and chunking
PDF Document	.pdf	25 MB	Text stream extraction, OCR fallback, structural paragraph parsing
Word Document	.docx	25 MB	OpenXML body parsing, header/table extraction

---

Rate Limiting Tiers

• Document Upload: RateLimitTier.EXPENSIVE — Protects background vectorization queues and object storage quotas.
• Document Listing & Download: RateLimitTier.READ — High-throughput cached metadata and file streaming.
• Document Deletion: RateLimitTier.WRITE — Fast purging across metadata, storage, and vector indexes.

---

Available Endpoints

Method	Endpoint	Description
POST	/api/v1/knowledge-base/upload	Upload a document to storage and vector database
GET	/api/v1/knowledge-base	List all uploaded documents for the organization
GET	/api/v1/knowledge-base/{document_id}/download	Securely download the raw document file
DELETE	/api/v1/knowledge-base/{document_id}	Permanently purge a document from DB, storage, and vectors

---

‹ Previous

Upload document

POST /api/v1/knowledge-base/upload

Next ›