Your clients are asking for AI
Build it for them on Certant, under your own brand, with models and prices you control, so a supplier's decision can't break what you delivered.
- Your client's name and colours on it
- The page each answer came from
- Running on your client's own hardware
Spot the request you're already hearing
Associations, firms and agencies want answers from their own documents. Most have no team to build it, and the regulated ones need it kept on their own infrastructure.
"Can our members get answers from our policies at 2am?"
You build a chatbot for their website that answers from their own policies and links to the page each answer came from.
"Can it check our agreements and flag what's out of line?"
You build an agent that reads each agreement against the client's own rules, flags the clause, and waits for a person to approve what happens next.
"Can I question our reports without waiting on the data team?"
You give them an analytics builder they can ask in plain English. It builds the chart from their own data and shows the source behind each number.
Each of these runs in Certant's cloud or on your client's own hardware. The cloud uses hosted models and bills usage in credits, while your client's hardware runs open-weight models under a fixed annual licence. A client can start in the cloud and move later without buying it again.
Count what a DIY build needs
Your clients may well wait for you rather than buy from a vendor they don't know. A head of procurement told MIT's researchers: "Our established BPO partner already understands our policies and processes. We're more likely to wait for their AI-enhanced version than switch to an unknown vendor."
MIT NANDA, The GenAI Divide: State of AI in Business 2025, preliminary findings from interviews and surveys.
All 22 have to work together before a client trusts one answer. On Certant they already do, so your team's time goes on the client's own documents and systems.
| Part of the build | What you would build | Systems |
|---|---|---|
| Reading documents | PDF, Word and slide parsers, OCR for scanned pages, and chunking that keeps tables whole | 3 |
| Understanding them | Entity and relationship extraction, a self-learning ontology, and a graph that updates as files change | 3 |
| Answering questions | Vector and graph search, structured queries over data, a retrieval router for each question, paragraph-level citations, and refusals when the answer isn't there | 5 |
| Products on top | A multilingual chat widget, an agent builder with approvals, analytics with 17 chart types, and monitors that raise alerts | 4 |
| Security | SSO and SCIM provisioning, per-document permissions, customer-held encryption keys, and an audit log of every prompt | 4 |
| Running it | GPU model serving that works offline, signed updates for air-gapped sites, and IRAP-aligned review evidence | 3 |
| Total | 22 |
Expect the model to change
On a model rented by the token, the lab decides when it retires. Anthropic promises at least 60 days' notice, and OpenAI three to six months depending on the model. A government order can come with none.
Each change means retesting the prompts, agents and answers your client has already signed off.
-
OpenAI gives developers three months' notice that its chatgpt-4o-latest model will leave the API. It goes on 17 February.
OpenAI deprecations -
OpenAI schedules a set of older models, including the first GPT-4o release and o1, for shutdown on 23 October.
OpenAI deprecations -
A US government export order leads Anthropic to switch off its two newest models for all customers, with immediate effect.
Anthropic's statement -
Access starts coming back, the day after the controls are lifted.
Anthropic's update
Keep the model you tested
Certant's sovereign tier ships with open-weight models such as Llama, Mistral and Qwen. The weights are files on your client's hardware, so the version you tested keeps running until you decide to move.
| What is compared | A model rented by the token | An open-weight model on your client's hardware |
|---|---|---|
| Price per question | Set by the lab, on its schedule. | Covered by a fixed annual licence. |
| When it retires | When the lab's notice period runs out. | When you choose to upgrade. |
| Who can switch it off | The lab, or a government order. | Your client. It runs on their hardware, offline if they want. |
| Where questions go | To the lab's servers. | Nowhere outside your client's network, unless they connect an outside system. |
Each open-weight model comes under its maker's licence. On Certant's cloud, answers come from hosted models that Certant rents.
Quote a number that holds
The Information, as reported by CNBC and Reuters, September 2025. US dollars.
The labs selling tokens today spend far more than they earn, and investors cover the gap. Prices set that way can rise once investors want a return, and a fixed-price client contract absorbs the difference.
On your client's own hardware there is no per-token bill. Your margin stops depending on a price list someone else can change.
| Line | How it's charged |
|---|---|
| Hardware | Bought once. It's either your client's own, or a Certant appliance that arrives with everything installed. |
| Licence | Twelve months, renewable, at a price agreed up front. |
| Usage | No charge per question. Every question runs on hardware that's already paid for. |
| Your services | Implementation, integration and support, priced by you. |
Clients in Certant's cloud pay in credits instead.
Choose how you build it
Your team builds on the same platform our own customers use, in the app or in code.
Set up knowledgebases, chatbots, agents, analytics and Strata without code. Or ask the AI FDE to build them from a plain-English request.
Add a chatbot to your client's website or member portal with one script tag.
Call Certant from your own software to add cited answers and agents to products you already sell or maintain.
Put your own brand, or your client's, on the chatbot and on the app their staff use.
Run a separate workspace for each client from one partner account. We set this up as part of an enterprise agreement.
Every answer shows the page it came from, so your client's staff can check it before acting.
What partners ask us first
Can we put our own brand on it?
Yes. The chatbot your client's members use and the app your client's staff work in can both carry your brand or your client's.
Do our clients need their own GPU hardware?
No. We can supply an ASUS Ascent GX10 with Certant already installed, which ASUS and Certant sell together. We can also install it on hardware or a cloud account your client already has, or run it in our cloud.
Are open-weight models good enough for client work?
Yes, because we designed Certant's harness so they don't have to match a frontier model. Before the model sees a question, Certant has found the right paragraphs, worked out how the client's records connect, and calculated any figures with a database query. The model reads what it's given and writes the answer with its source, and it says so when the documents don't hold one. In work grounded in documents, finding the right paragraph does more for the answer than the size of the model.
We test open-weight models on our own hardware and publish the results. Read how three local model stacks compared.
Will Certant sell to our client directly?
Register the deal and we protect it for 90 days while you work it. We don't contact your client.
How long does a deployment take?
The appliance is usually running within a week of delivery. On hardware your client already has, it takes from the same day with the install wizard to a few weeks if there's integration work.
How is it priced, and what's our margin?
On your client's own hardware, the cost is the hardware plus a 12-month licence, quoted for each deployment. Cloud clients pay in credits. Your margin grows with the depth of the deployment, and we go through the numbers on the call.
ASUS and Integr8tiv already deliver Certant
Certant is in production with unions and associations in Australia and the UAE, with more than 120,000 pages processed so far.
ASUS sells a sovereign Certant on its Ascent GX10
We worked with ASUS to build a sovereign version of Certant that runs on its Ascent GX10 machines. ASUS and Certant sell it to customers together.
Integr8tiv deploys Certant for member organisations
Integr8tiv connects Certant to the systems associations and mutuals already run, such as Xero, Mailchimp, HubSpot and Salesforce. It supports them after go-live, including on sovereign deployments in Australia and the UAE.
Bring us the client who asked first
Tell us which clients you serve and we'll show you what a fixed-price, self-hosted deployment looks like for each of them.
What happens next
- You send the form. Your name, email and firm are all we need.
- A person on the team replies within one business day, Australian hours, to set a time.
- On the call we work out how to run Certant for each client, what it costs and how long it takes to install.
What we commit to
- We don't contact your client.
- Register a deal and we protect it for 90 days.
- Implementation, integration and support revenue stays with you.
- Your engineers get a direct channel to the people who wrote the platform.