AI Agent Development
Build assistants around your knowledge and tools, with review and controls suited to the task.
- LLM APIs
- RAG
- Node.js
- Next.js
Overview
An AI assistant is useful when it is grounded in your own material — your documents, products, policies or records — rather than answering from general knowledge. This service covers connecting a model to that material and to the tools it needs to act.
How much the assistant is allowed to do on its own is a scoping decision, not a default. Anything that touches a customer, a payment or a record is built with a human approval step unless you decide otherwise.
Problems this solves
- Staff time goes to questions already answered in a document
- Information is spread across files nobody can search
- A generic chatbot gives answers that are wrong for your business
- Routine requests queue behind work that needs a person
Who this service suits
- Teams answering the same questions repeatedly from known material
- Businesses with documentation customers cannot navigate
- Support functions handling high volumes of routine enquiries
- Companies wanting internal staff assistance over their own systems
Features and deliverables
Grounding
- Connecting the assistant to your documents and data
- Retrieval setup so answers cite your own material
- Handling for questions it should refuse or escalate
Action and control
- Business tool and API connections
- Human review and approval steps before an action
- Access control over what the assistant can see and do
- Logging of conversations and actions taken
Delivery
- Chat interface on your website or inside an internal tool
- Evaluation against real questions before launch
- Handover with guidance on monitoring and limits
Technologies we use
The final stack is chosen per project, based on scope, budget, who will maintain the site and what it needs to connect to. Not every technology listed is used on every project.
AI layer
- LLM APIs
- The language model behind the assistant, accessed through a provider API and billed by usage.
- Retrieval (RAG)
- Feeds the model your own documents at question time, so answers come from your material.
Application
- Node.js
- Runs the services, tool connections and access control around the model.
- React.js and Next.js
- Build the chat interface and any admin screens for reviewing conversations.
How a project runs
The same stages apply to every project. Which of them are included, and how long each takes, is set out in your proposal.
- 01
Requirement Discussion
We go through what you need, who it is for and which features matter most, so the scope is built on your actual priorities.
- 02
Scope and Proposal
You receive a written proposal setting out the deliverables, assumptions, timeline and cost before any work begins.
- 03
UI/UX and Planning
Structure, screens and user journeys are designed and agreed, along with the technical approach for the build.
- 04
Development
The approved scope is built in stages, with the work shared as it progresses rather than only at the end.
- 05
Testing and Quality Assurance
Functionality, responsiveness, performance and content are checked across devices and common browsers.
- 06
Client Review
You review the build against the agreed scope and raise changes, which are handled before deployment.
- 07
Deployment and Handover
The project goes live on your hosting or platform account, with access, credentials and a walkthrough handed over.
- 08
Post-Launch Support
Where support is included in the agreement, the covered period and scope are stated in the proposal.
Timeline
A delivery estimate is given in the written proposal, once the scope is agreed. Estimates are not delivery guarantees: the date moves with the scope and with how quickly reviews and content come back.
What affects the timeline
- The volume and state of the knowledge being connected
- The number of tools or systems the assistant must reach
- How much review and evaluation the use case requires
- Approval turnaround on answer quality
How pricing works
This service is quoted per project. Tell us what you need and we will come back with a written proposal covering scope, deliverables, timeline and cost.
A project quotation is separate from the monthly care plans, which cover ongoing changes after launch.
Add-ons and what is not included
Optional add-ons
Quoted separately, on top of the agreed scope.
- Additional knowledge sources or tool connections
- Ongoing evaluation and prompt tuning
- Multi-language support
- Voice or telephony channels
Third-party costs and exclusions
Paid directly to the provider, or outside the project scope.
- AI model provider usage charges, which are metered and billed to you
- Hosting, vector database and infrastructure costs
- Third-party subscription charges
- GST, where applicable, is charged additionally
Start a project enquiry
Let’s talk about your
AI Agent Development project.
AI Agent Development is already selected below. Send a few details and we will come back with next steps and a written proposal — you don’t need a technical brief to start.
- What you want to build or improve
- The tools or systems you already use
- Your timeline and budget, if known