Agencies selling AI assistants face a predictable challenge: every client wants their brand, their docs, and zero cross-client data leaks. Oprag’s project model maps cleanly to client engagements — upload their corpus, validate cited answers, embed or API-integrate under their domain.
One project per client
Create a separate Oprag project for each client engagement. This gives you:
- Isolated document stores and API keys
- Separate golden question sets per client
- Clean handoff when the engagement ends (archive or export)
Never mix client PDFs in a shared project — support and compliance reviews get painful fast.
Workflow from kickoff to launch
Discovery (week 1)
- Collect canonical docs: policies, FAQs, product guides
- Identify questions that must never be auto-answered (legal, medical, regulated advice)
- Agree on citation UX — inline links vs sidebar sources
Validation (week 1–2)
Run the client’s top 20 support themes in the dashboard. Client stakeholders sign off before embed code ships. Use validate document Q&A before launch.
Integration (week 2)
Pick delivery mode:
- Widget embed for marketing sites and help centers — embed guide
- REST API for custom React/Vue apps — integration guide
Style the widget with client CSS variables; keep their logo and tone in surrounding chrome.
Handoff (week 3)
Deliver:
- Admin access or documented upload process for doc updates
- Golden question doc they can re-run after content changes
- API key rotation instructions
Pricing and packaging for clients
Common agency packages:
- Setup fee — ingestion, golden set, embed
- Monthly retainer — doc updates, monthly quality review
- Usage pass-through — align Oprag Pro tier with expected query volume
Be explicit that infrastructure runs on Oprag’s AWS with tenant isolation — not inside the client’s AWS account unless they separately integrate via API.
White-label checklist
- Custom domain embed (client subdomain)
- No Oprag branding in client-facing UI (your embed wrapper)
- Privacy language in client contract matches security page
- Separate API keys per environment (staging vs prod)
Scale across many clients
Template your onboarding Notion doc: upload checklist, golden questions, sign-off form. Most margin comes from repeatable validation, not custom RAG engineering.
For multi-tenant SaaS clients instead of agency one-offs, see tenant-isolated RAG for B2B SaaS.