A · Generic chatbot
Knows the internetReturn windows are usually between 14 and 30 days, but it depends on the supplier. You may want to check your agreement.
No sources · a confident guess
AI that knows your business.
Private business intelligence
Private AI built on your own documents, data and systems. It answers your team with sources, and only from what you approve.
(02)How it works
Knowledge in. Answers out.
-
01
Connect
We connect the sources you approve: documents, databases, your CRM, the know-how your team already wrote down.
-
02
Retrieve
When someone asks, the system finds the passages that answer it and hands them to the model as evidence.
-
03
Answer
The answer goes back in plain language, with its sources, to the person who asked, if they are allowed to see it.
Illustrative
(03)The difference
Not a chatbot with your logo on it.
A generic chatbot knows the internet. A custom AI knows your business, shows where each answer came from and checks who is asking.
Same question, asked twice
“What's our return window on wholesale orders?”
Example · fictional documentsB · Custom AI
Knows your business30 days for unopened stock1. Opened items can come back within 14 days, as store credit only2.
- 1Wholesale T&Cs §6.2
- 2Returns Policy v4, p.4
2 sources · permission checked · logged
| Aspect | Generic chatbot | Custom AI |
|---|---|---|
| 01What it knows | The public internet, up to its training date | Your documents, data and systems, kept current |
| 02Where answers come from | A plausible guess, no sources shown | Passages from your approved sources, cited |
| 03Who can ask what | Anyone, about anything | Follows your roles and permissions |
| 04What gets recorded | Nothing you control | Every question, source and answer, logged |
| 05Over time | Stays generic | Gets sharper as your information grows |
(04)What we build07 solutions
Seven solutions. One private foundation.
Every build sits on the same layers. What changes is the part your team touches.
01 Private AI Assistants
An assistant your team can ask about anything in the business. It answers from your documents and shows where it found each answer.
- Layers
- SourcesRetrievalModelGuardrailsInterface
- Good for
- Support, sales and onboarding questions
02 RAG Systems
The engine underneath: your content indexed, searched and handed to the model as evidence, so answers stay grounded in what you actually wrote.
- Layers
- SourcesRetrievalModelGuardrailsInterface
- Good for
- Grounding any AI feature in your own content
03 Knowledge Platforms
One home for policies, guides and know-how. Organised, searchable and kept current as documents change.
- Layers
- SourcesRetrievalModelGuardrailsInterface
- Good for
- Policies, playbooks and internal wikis
04 AI Dashboards
Your numbers, explained. Dashboards that say what changed and why, in plain language, next to the chart.
- Layers
- SourcesRetrievalModelGuardrailsInterface
- Good for
- Weekly reporting and leadership reviews
05 Internal Search
Search that understands the question, not just the keywords, across every approved system at once.
- Layers
- SourcesRetrievalModelGuardrailsInterface
- Good for
- Finding the right file, clause or record fast
06 Data Intelligence
Patterns, outliers and summaries pulled from your operational data and sent to the people who act on them.
- Layers
- SourcesRetrievalModelGuardrailsInterface
- Good for
- Orders, stock, pipeline and service data
07 Custom Applications
AI built into your own tools: a quoting helper, a document checker, a portal feature. Shaped around the job, not a template.
- Layers
- SourcesRetrievalModelGuardrailsInterface
- Good for
- Workflows no product covers out of the box
- L5 Interface Where your team uses it
- L4 Guardrails Roles, permissions, audit trail
- L3 Model Reasoning over the evidence
- L2 Retrieval Indexing, search and ranking
- L1 Sources Your approved documents and data
(05)Try it
Ask your business.
Pick a question. Watch the answer come from the documents, not from a guess.
Choose a question
Question: What's our return window on wholesale orders?
Answer: 30 days for unopened stock1. Opened items can come back within 14 days, as store credit only2.
- 1Wholesale T&Cs §6.2
- 2Returns Policy v4, p.4
(06)How we build it
Private by design.
Guardrails aren't a feature we bolt on at the end. They are how the system is built from the first day.
-
Approved sources only
The AI reads what you connect and nothing else. You choose every source, and you can remove one at any time.
-
Access follows your roles
People only get answers from documents they're allowed to open. Finance stays with finance.
-
A record of every answer
Every question, source and answer is logged, so you can see what was asked and why it answered the way it did.
-
People stay in the loop
Sensitive answers can go to a person for review, and their feedback makes the next answer better.
01 Sources
- Wholesale_T&Cs.pdf
- Onboarding_Guide_v3.docx
- Orders database
- CRM · Accounts
- Team knowledge base
- Shipping_Matrix.xlsx
- Returns_Policy_v4
- Product catalogue
- Public web · not connected
02 Access
| Role | Pricing | Policies | Payroll |
|---|---|---|---|
| Sales | Allowed | Allowed | No access |
| Support | No access | Allowed | No access |
| Finance | Allowed | Allowed | Allowed |
03 Audit trail
- SupportReturn window, wholesale2 sources · answered
- SalesMUG-11 to Canada2 sources · answered
- SupportPayroll datesBlocked · no access
- OpsWeather tomorrowOut of scope · logged
04 Review
Can we offer 60-day terms to a new wholesale account?
Commercial terms · routed to a manager
(07)Where to start
Start with one question.
You don't need a data strategy deck to begin. You need the question your team asks most, and the documents that answer it.
-
01
One question
Pick the questions that cost your team the most time to answer.
-
02
A few sources
Connect only the documents and data that answer them.
-
03
One team
Pilot with the people who ask most, and watch what changes.
-
04
Then widen
Add sources, teams and tools as the answers earn trust.
Project stories go here once clients approve them for publication. Until then, ask us to walk you through how a build works.
(08)Start here
What will your AI know?
Tell us the questions your team keeps asking. We'll show you what a private AI built on your own information could answer.