Knowledge Base (RAG)

Retrieval-Augmented Generation (RAG) lets J.O.S.I.E answer from your PDFs and documents — not just the system prompt.

Available on Pro and Enterprise plans.

Limits by plan

Basic and free tiers cannot upload documents.

Supported formats

• PDF (recommended)
• DOCX (converted to text server-side)
• Plain text exports

PDFs with clear headings and Q&A structure retrieve best.

Upload flow

1. Go to Dashboard → Documents
2. Click Upload or use the Knowledge Wizard
3. File is stored in tenant-isolated storage
4. Text is extracted, chunked, and embedded (OpenAI embeddings)
5. Status shows Indexed when ready

Knowledge Wizard (recommended)

The wizard builds an RAG-optimized PDF from your inputs:

1. Business basics — name, industry, contact
2. Products / services — what you offer
3. FAQs — question/answer pairs with keywords
4. Policies — shipping, refunds, privacy snippets
5. Review & generate — produces structured PDF, auto-uploads and indexes

Wizard modes

Wizard metadata is saved so you can edit and re-export later.

How retrieval works

When a user asks a question:

1. Query is embedded and matched against your chunk vectors (pgvector)
2. Top chunks within distance thresholds are injected as context
3. The model answers only from provided context when guardrails require it
4. If no relevant chunk is found, J.O.S.I.E says it doesn't have that information (no hallucination)

Pro/Enterprise use tier-specific distance thresholds so legitimate FAQ wording still matches.

Writing content that retrieves well

Structure documents as:

QUESTION: How long do withdrawals take?
ANSWER: Withdrawals are processed within 24–48 business hours after approval.
KEYWORDS: withdrawal, payout, processing time, pending

Tips:

• One topic per block — avoid 10-page walls of text
• Use customer phrasing in QUESTION lines
• Repeat keywords naturally in ANSWER and KEYWORDS
• Update via wizard or upload new version; re-index replaces old chunks for that doc

Download documents

From Documents, download the original file for backup or compliance.

Static knowledge vs. RAG

Use both: static knowledge for hours/support email; RAG for product specs and policies.

Troubleshooting RAG