Setup: RAG Search Engine
How to get semantic search running on your documents — from first contact to live deployment.
Prerequisites
- ✓ Your documents in any format your team already has them in
- ✓ A paid plan (RAG Search is bundled free) — or a standalone engagement
- ✓ Nothing technical required on your side
What You Provide
Send your documents in any format. Examples:
| Document type | Examples |
|---|
| Operational documents | Call scripts, clinic intake guides, property listing docs, onboarding materials |
| Policy documents | HR and leave policies, compliance procedures, employee handbooks |
| Product data | Product catalogs, pricing sheets, service descriptions |
| Knowledge base | FAQ documents, knowledge base articles, training manuals |
| SOPs | Standard operating procedures, escalation guides, process documentation |
| Legal documents | Contracts, compliance documents, terms and conditions |
Accepted formats: PDF, DOCX, TXT, CSV, JSON, spreadsheet exports. Any format your team already uses.
Setup Steps
- • Send 100 records for a free test — or your full corpus for deployment
- • Use the contact form or send directly to your deployment partner
- • No reformatting needed — send them as they are
- • Documents are chunked using a recursive character text splitter
- • Each chunk is embedded as a 1,536-dimension vector
- • Stored in a pgvector HNSW index for sub-50ms retrieval
- • Multi-tenant by design — each client's documents tagged to their API key
- • We run real queries against your data and send back results
- • You see exactly what your users would see
- • Free 100-record test: results returned within 48 hours
- • Adjust and refine before going live
- • Search interface built to your branding (standalone track)
- • Or integrated into Maya across all your channels (bundled track)
- • API key scoping and rate limiting configured
- • Web front end hosted — no infrastructure on your side
Two Deployment Tracks
Track 1 — Bundled With Your Paid Plan
If you are on any paid plan, RAG Search is already included at no extra cost.
- • Your deployment partner handles document submission as part of onboarding
- • Documents loaded during deployment — nothing technical on your side
- • Co-pilot mode: Maya searches your docs in real time and surfaces answers on screen
- • Works across all active channels — website chat, messaging, voice
- • 30 languages — query in one, return results from any
Extra cost: $0
Track 2 — Standalone (Search Only)
For organisations that only need document search — no voice, no screening tools.
- • Corpus ingestion + embedding
- • pgvector deployment + HNSW index
- • Search interface built to your branding — name, logo, colours, result template
- • API key scoping + rate limiting
- • Hosted infrastructure
- • Larger customisations (multi-language UI, custom scoring, API integrations) quoted separately
How It Works
| Step | What happens |
|---|
| Document ingestion | Your documents are split into chunks using a recursive character text splitter with 1,000-character chunks and 200-character overlap |
| Embedding | Each chunk is embedded as a 1,536-dimension vector — state of the art for semantic retrieval |
| Indexing | Vectors stored in a pgvector HNSW index for sub-50ms cosine similarity search |
| Query | User's question is embedded into the same vector space, then matched against your documents by meaning |
| Result | The relevant excerpt from your document is returned — with the source file cited. Nothing invented. |
Multi-tenant architecture: Each client's documents are tagged to their API key. Mandatory filter on every query — impossible to access another client's data. Your documents are isolated and secure.
Adding New Documents
- • No reindexing delays — add a document and it is searchable in seconds
- • Send new documents to your deployment partner or upload through your TaskMaster Agent
- • Embedding cost per document is negligible — under $5 for a typical corpus
- • Monthly search cost is under $0.50 at typical business volumes
What RAG Search Does Not Do
- • It does not guess or generate answers from general knowledge
- • It does not invent information that is not in your documents
- • It searches only your corpus, returns only your content, and cites your source files
- • If the answer is not in your data, nothing is returned — no hallucination
Free Test
Send 100 records in any format — CSV, JSON, PDF, or a spreadsheet export. We index them, run sample queries, and send back a demo within 48 hours. Free. No commitment. You see exactly what your users would see before agreeing to anything.