Agent Setup is KitelyTech's fixed-scope service for designing, building, and deploying one production AI agent, from scoping through handoff. An organization that needs a working AI agent (customer intake, document Q&A, a defined workflow) often lacks the platform-specific knowledge to design, guardrail, and deploy it safely.
Agent design and scoping: what it does, what it doesn't, and its edge cases
Platform selection, matched to the client's existing architecture and requirements
System prompt engineering: guardrails, tone, token limits, and resistance to manipulation attempts
One integration: a knowledge base, CRM query, form, or messaging platform connection
Testing against 20+ client-provided scenarios and failure modes
Handoff documentation and a defined period of reactive support
Additional integrations beyond the first: scoped separately
Data preparation for a source that isn't already clean and structured: see Data Integration & Structuring
Custom monitoring dashboards
Compliance frameworks beyond the standard template
Price: fixed for the defined scope, set during scoping.
Guardrails reduce the likelihood of unintended behavior. They are not a guarantee of it, and we describe them that way in every engagement.
Get it right the first time: in Moffatt v. Air Canada, a Canadian tribunal held the airline responsible for inaccurate travel-policy information its website chatbot gave a customer, awarding damages after the airline argued the chatbot's statements weren't legally its own. The tribunal wasn't persuaded; it treated the bot as no different from any other page on the company's site. The lesson generalizes well beyond airlines: an agent that speaks with company authority creates company liability the moment it's wrong, which is why testing against documented failure modes is a fixed part of this engagement, not an optional add-on.
Often yes, and for a low-stakes internal experiment you probably should. This service exists for agents that will speak or act with your company's authority, where guardrails, edge cases, and failure testing are the difference between a tool and a liability.
The company is, and courts have said so; the Air Canada tribunal decision described above is the reference case. That's why testing against documented failure modes is a fixed part of this engagement rather than an optional add-on.
One connection to a knowledge base, CRM query, form, or messaging platform. Additional integrations are scoped separately, and if your data source needs cleaning or retrieval setup first, that's the Data Integration & Structuring service.
No, and anyone who tells you otherwise is selling false safety. Guardrails reduce the likelihood of unintended behavior. We describe them that way in every engagement, and we test against client-provided scenarios so the remaining risk is understood rather than hidden.
