*Last updated: May 2026*
Linkki is a brand and technology practice based in central TƒÅmaki Makaurau (Auckland), working with clients throughout New Zealand and beyond. Our primary offering is custom retrieval-augmented generation (RAG) systems that put a business’s own information to work. We build these under a method we call BrandRAG, which puts strategy and content architecture before any engineering. Alongside the RAG practice, we also run a self-service brand strategy platform that connects to ChatGPT and Claude.
We build knowledge systems for organisations where a wrong answer has consequences. That includes professional services with dense document libraries, regulated industries where every answer is a compliance statement, trade and industrial businesses with specialist product knowledge, and internal knowledge teams where a confident wrong answer is worse than no answer. Example clients include AFH Forklifts, Tāwhia Copyright Aotearoa, and Accurate Consulting.
Three things separate Linkki from a standard RAG build. First, the order of the work: architecture before engineering, every time, with tool selection happening last. Second, the structure of the knowledge: information is organised so reliability is the dominant signal, not similarity, so the most trusted answers surface first. Third, the strategy layer: knowledge design and systems design make architectural decisions together before anything is built. Two senior practitioners lead each engagement, one for each discipline.
RAG stands for retrieval-augmented generation. The system finds relevant material from your own information, then uses it to answer a question, check a document, or generate something new, grounded in what your business actually knows. It is how AI retrieves your specific information instead of relying on what it learned from the internet.
A live AI-powered chatbot that answers user queries by searching your own document library rather than relying on general AI knowledge. It runs on your website through a custom JavaScript widget, finds the most relevant information from your documents when a user asks a question, combines that with your brand guidelines, and generates a response that is accurate and on-brand. Behind the widget sits an ingestion pipeline (which processes your documents) and a live query pipeline (which runs in real time on every user interaction).
BrandRAG is the method Linkki uses to build every RAG system. Most RAG solutions start with the technology. BrandRAG starts before that, with the brand strategy and content architecture that make the technology work. The method runs in three phases: a strategic review to capture what the business actually knows (including the knowledge that lives only in senior staff’s heads), an architecture phase to structure and prioritise that knowledge so the most reliable answer surfaces first rather than the most linguistically similar one, and an engineering phase that applies standard tools to above-standard preparation.
A model can only work with what it can find, and what it finds depends entirely on how the underlying knowledge has been organised. Standard retrieval finds information by similarity, which works when the right answer happens to be the most similar content. It fails when the most similar content is wrong, out of date, written for a different audience, or one of three competing versions of the same answer. Architecture is what fixes that, and it has to happen before anything is built.
Every engagement starts with a conversation, not a proposal. The work itself runs in three phases: strategic review, architecture, and engineering. Each phase has clear deliverables, a defined timeline, and a moment where the client signs off before the next phase starts. By the time we are building, the client knows what they are getting and why. After delivery, we stay close for ongoing maintenance, content updates, and adjustments as the business evolves.
Anywhere a business has a body of specialist knowledge that is too large for staff to memorise, too varied for generic AI to handle, and too important to get wrong. Common applications include sales assistants trained on a full product range, public-facing knowledge systems that answer queries grounded in regulation, document-checking systems that flag compliance issues before documents go out, and internal knowledge bases that surface trusted answers for staff.
AFH Forklifts is a nationwide forklift sales, lease, and hire company. We built a sales assistant trained on the full AFH range that answers buyer questions instantly and guides them toward the right machine and the right sales rep. Tawhia Copyright Aotearoa is the national copyright licensing body. We built a public-facing knowledge system that answers rightsholder and licensee queries in plain language, grounded in New Zealand copyright law. Accurate Consulting is an independent asbestos consultancy. We built a document-checking system that flags compliance issues in survey reports before they reach clients, grounded in New Zealand asbestos regulations.
It depends on the size and shape of the document library and how much elicitation is needed to capture the knowledge that has not been written down. A typical BrandRAG project takes around 2-3 months. Each phase (strategic review, architecture, engineering) has a defined timeline agreed before work starts, and the client signs off at the end of each phase before the next one begins. We can give you a realistic timeline after the first conversation.
In plain terms: when someone asks the chatbot a question, the system looks through your documents, finds the parts that matter, weighs them up against what your business considers most trustworthy, and uses that to write an answer in your voice. The user sees a simple chat experience on your website. Behind that is the architecture work we do during the build, which is what makes the answers reliable rather than just plausible.
We process your documents into a form the system can search and reason over. That includes cleaning them up, breaking them into manageable pieces, tagging them with information about what they cover and how trustworthy they are, and indexing them so the chatbot can find the right material quickly. The work is done with care: nothing important is summarised away, and verified human interactions (such as Q&As, email threads, and FAQs) are treated as the highest-priority source material.
This is where Linkki’s architecture earns its keep. Every piece of content in the system is tagged with how much it should be trusted, based on where it came from and how well it has been verified. When a user asks a question, the system pulls the relevant material and gives more weight to the most trusted sources before generating an answer. So a verified Q&A will outweigh an old supporting document, even if the supporting document happens to be worded more similarly to the question. Reliability beats similarity. That is the BrandRAG difference in one sentence.
Each Linkki RAG system is configured with a client-specific brief that defines tone of voice, audience, compliance boundaries, and what the chatbot should and should not say. That brief sits underneath every answer the system generates, so responses sound like your business rather than like a generic AI assistant.
Through a custom-built widget that sits on your website. Users see a chat window, type their question, and get an answer in your brand voice. The widget is part of the build, not a third-party plugin.
Every question, every answer, and every step the system takes in between is logged and reviewed. We watch for patterns: questions the chatbot is struggling with, gaps in the source material, retrieval that needs tuning, and anywhere the answers could be sharper. That review feeds directly into ongoing improvements, which are part of the monthly retainer.
It can answer questions grounded in your documents, draft responses in your brand voice, point users toward the right person or product, and flag when it does not have a confident answer. What it deliberately does not do is make things up. The system is configured to answer only from the material we have given it and to say so when the available information is limited or low-confidence. That is the whole point: a confident wrong answer is worse than no answer.
Our systems can be hosted on sovereign trans-Tasman infrastructure. The brand strategy platform’s data is stored in a Google Firebase environment on Australian servers. For RAG builds, hosting is configured per engagement, and we work within New Zealand and Australian privacy law throughout. Data sovereignty is treated as a first-order concern, not an afterthought.
We sanitise sensitive information before any AI processing, choosing the right tool for each situation. Where the situation calls for it, sanitisation can run on the client’s own systems with no internet connection, so nothing sensitive ever leaves their building. All Linkki systems are built to Privacy by Design principles and are compliant with New Zealand and Australian privacy law.
Linkki Wash is a planned proprietary local LLM service that will strip sensitive information from client files before they enter the RAG pipeline. The plan is for it to run locally so sensitive data does not leave the client’s environment. Linkki Wash is not yet live. When it is available, it will be scoped and priced separately from the standard RAG build and retainer.
The system needs to read your documents in order to answer questions about them, so yes, the AI sees the relevant material at the moment it generates an answer. What we are careful about is what counts as “sensitive”. Anything genuinely sensitive is sanitised out before your documents enter the system, either through our standard process or, when Linkki Wash is live, through a local pass that never leaves your environment. The system is designed so the AI sees what it needs to answer the question, and nothing more.
A new build is a separate project engagement, scoped and priced based on the size of the document library, the complexity of the elicitation work, and the integrations required. We give you a fixed price after the first conversation and the initial scoping work. A typical BrandRAG system costs around $15-25k (NZD). The build is distinct from the ongoing managed service retainer, which kicks in once the system is live.
The retainer starts from NZD $800 per month, all-inclusive. That covers infrastructure (hosting, vector database, LLM and embedding usage, monitoring) and management (document and pipeline management, monitoring and reporting, performance tuning). Larger document libraries, higher chatbot volumes, or more involved management needs may push the retainer higher. We agree the figure with you upfront based on your specific system, and there are no surprise line items.
Document maintenance (adding, removing, and updating sources as your content evolves, and re-processing them so the chatbot stays current), system health checks across the live infrastructure, monitoring and reporting (with a monthly summary), tuning to improve answer quality over time, priority support with response within one business day, infrastructure oversight (keeping an eye on costs and usage so there are no surprises), and quarterly reviews covering performance, document coverage, and recommendations for improvement.
Rebuilding or reconfiguring the chatbot (changes to the underlying system, the website widget, or how the architecture is set up), onboarding new clients to the platform (a new RAG system from scratch is a separate project engagement), custom development work (new features, integrations, or significant changes are quoted on a time-and-materials basis), and Linkki Wash (not yet live, will be scoped separately when available).
The retainer starts from NZD $800 per month, invoiced at the start of each month, with all infrastructure costs included. Notice period is 30 days written notice to end the retainer. If chatbot volume grows significantly (for example, 10x current usage), infrastructure costs are reviewed and any increase is communicated before being passed on. Work outside the retainer scope is agreed in writing before commencing.
A self-service tool that helps businesses build a guided brand strategy in seven steps and connect it to ChatGPT and Claude through a secure MCP connector, so AI outputs stay on-brand. It was Linkki’s original product. It is now a secondary offering, with the primary focus on bespoke RAG builds. The platform continues to be developed and supported for its existing customer base.
Two groups. Leaders managing teams whose staff are producing content and communications with AI tools every day, where without a shared strategic layer underneath the output ends up inconsistent, off-brand, and sometimes off-message in ways that matter. And business owners doing their own marketing, using AI to write content, run campaigns, and stay visible, who need a clear brand strategy connected to their tools so the output sounds like their business rather than everyone else.
$25 per month, with a 3-month minimum at the start, then month-to-month. Cancel anytime after that. Optional add-ons include a 1-on-1 brand strategy consultation with a Senior Brand Strategist for $650, and brand and logo design from $1,400.
Through a secure MCP (Model Context Protocol) connector with OAuth authentication. Once your brand strategy is built within the platform, that strategic context connects to ChatGPT and Claude so first drafts come out aligned with your positioning, voice, and compliance rules.
The first step is a conversation, not a proposal. We look at what your business knows, where the value is, and what the right approach is to solve the challenge. No slide decks, we promise (unless requested). From there, if a RAG build is the right fit, we move into a scoping engagement. If something else makes more sense, we will say so.
Our studio is in central Tāmaki Makaurau (Auckland) and we work with clients throughout New Zealand and beyond. The practice operates entirely within New Zealand and Australian privacy law.
Two senior practitioners lead every engagement. Michael Easton, Co-Founder and Strategist (Knowledge design), has thirty years in strategy and communications across government, enterprise, and professional services. He runs elicitation, content architecture, and the brand-voice layer of every BrandRAG engagement. Pekka Malkamäki, Co-Founder and Strategist (Systems design), has spent his career leading teams building digital platforms and the infrastructures behind them. He leads the technical design of every BrandRAG engagement and oversees the team that manages and optimises each system on an ongoing basis.
Linkki is a brand and technology practice based in central Tāmaki Makaurau (Auckland), working with clients throughout New Zealand and beyond. The practice operates entirely within New Zealand and Australian privacy law.
Linkki is built around two strategy disciplines, knowledge design and systems design, both led by senior practitioners who shape each system before anyone starts building it.
Brands using Linkki
Or give us a call on +64 21 280 2773
Or give us a call on +64 21 280 2773