# Technical Brief: AI Agent Setup in respond.io ## You+Baby Studio — Photography Studio, UAE & Qatar You+Baby Studio is a premium family photography studio operating in Dubai, Abu Dhabi, and Doha. We use respond.io as our CRM and messaging platform (WhatsApp, Instagram, Facebook, Email). We currently have an AI agent system in respond.io that partially works but has reliability issues. We need someone to fix, rebuild, or significantly improve it. You will be provided with all existing flows, prompts, and knowledge base files, plus fully updated versions to work from. --- ## What We Need A fully functional two-agent system inside respond.io with supporting Workflows and a clean, maintainable knowledge base. ### Agent 1 — Info Collector Handles the very first part of the conversation. Collects four things: - Client language (detected automatically from first message) - Client location (country + city/area) - Session type (newborn, maternity, cake smash, etc.) - Baby's age or pregnancy weeks (where applicable) Once all three are confirmed, Agent 1 sends a closing message and stops completely. It does not answer questions, does not send prices, does not engage further. ### Workflow (between agents) Triggered automatically when Agent 1 finishes. Based on session type + city, it sends the correct package image to the client, then hands the conversation to Agent 2. ### Agent 2 — Consultant & Booking Closer Picks up after the package image is sent. Answers client questions (packages, prices, inclusions, themes, home service, delivery, safety, etc.), does a light upsell, collects a preferred date and time, then assigns to the human reservations team. --- ## Known Issues to Fix The previous system had the following recurring problems — these must be addressed: 1. **Package images not always sent** — the workflow sometimes skips the image step 2. **Chat not assigned to Agent 2** after the image is sent — clients get no follow-up 3. **Agent going off-script** — answering questions it shouldn't, inventing information, or ignoring the hard stop instruction 4. **Knowledge base hard to update** — prices and package details were embedded in the prompt, making any change require editing the whole prompt 5. **Location detection failures** — agent didn't recognize areas, districts, or Arabic spellings of cities --- ## Deliverables 1. **Agent 1 prompt** — updated and tested in respond.io 2. **Agent 2 prompt** — updated and tested in respond.io 3. **Workflow** — trigger on Session Type field → send correct image by session type + city → assign to Agent 2 4. **Knowledge base** — clean, structured file(s) that we can edit ourselves without touching the prompts. Must cover: all session types, packages, prices by city (Dubai / Abu Dhabi / Doha), extra services, home service fees, delivery timelines 5. **Short guide** — how to update prices or package info going forward (which file to edit, how to upload it to respond.io) --- ## What You Will Receive From Us - Current respond.io flow and prompts (for reference) - Fully updated Agent 1 prompt (ready to use or adapt) - Fully updated Agent 2 prompt (ready to use or adapt) - Clean pricing knowledge base (structured markdown) - Session type & wardrobe links knowledge base - Package images (per session type and city — we will provide or clarify) - Access to respond.io workspace for setup and testing --- ## Key Requirements - Must be built natively inside **respond.io** (AI Agent feature + Workflows) - Knowledge base must be a **separate file** (not embedded in the prompt) so we can update prices without touching the agent logic - Must handle **English and Arabic** — language detected automatically from client's first message - Location detection must cover **city districts and area names**, not just country/city (e.g., "Marina", "Khalifa City", "Al Gharrafa" should all map correctly) - Agent 1 must have a **hard stop** — once it sends the final message, it must not respond to anything further - Agent 2 must write an **internal note** with session details before every handoff to the human team - Both agents must follow strict communication rules (no forbidden words, no invented information, WhatsApp tone) - **Timeout logic** must be built in — if client goes silent, agent sends a follow-up after a set time and assigns to human ## How to Apply Please share: 1. Your experience with respond.io specifically (AI Agents + Workflows) 2. Any similar projects you've done (chatbot or CRM automation for service businesses) 3. Your suggested approach — fix existing flow or rebuild from scratch, and why 4. Estimated timeline and cost We are happy to jump on a short call to walk you through the existing setup before you quote.