One platform to build and run AI workers.
Our AI Platform is the tested core under every AI worker we build. It runs on your servers or in your region. Most AI workers on it go live in 4 to 6 weeks, and they are affordable to run.
It starts with a short call with our AI experts.
In short
Our AI Platform is the tested core under every AI worker we build. It is for large companies that need AI on their own servers or in their region. Every step is logged, and you pay a fixed licence, not a bill per question.
- 7 layers, 26 tested modules
- Your servers or your region
- Every step logged
- A fixed licence
The AI runs on your servers.
Regulators, auditors and your own customers ask where your data goes. Our answer is short: on your servers, or in your region.
How we keep it safeThe AI model is only one part.
Anyone can rent a smart AI model. On its own, it cannot do the six jobs below. Our AI Platform does them for every AI worker.
01 · What you can rent02 to 07 · What we built around it
- 01
The AI model
Reads and writes language. Many companies sell one.
What you can rent
- 02
Find the right fact
Searches years of messy files, even scans and photos.
- 03
Check who may see it
Shows each person only what they may see at the source.
- 04
Prove where it came from
Each answer links to the document it came from.
- 05
Follow your rules
Fixed checks give the same answer every time.
- 06
Ask a person first
Risky steps wait for a named person to approve.
- 07
Keep a full record
Every step is saved, so an auditor can replay it.
- All seven together
A working AI worker
Our AI Platform is that engineering, built once and reused.
See the seven layers
Same core. Only your setup changes.
The core is the same for every customer. Only your setup changes, so new AI workers are quicker to build.
Warranty Assist, in production.
Claims, invoices, service records and photos.
Reused evidence found, mismatches caught, policies checked and a risk worklist.
01
AI Platform core
Same for every AI worker
Connectors, ingestion, OCR and CV
Reads your systems and files, even scans and photos.
Hybrid search, graph and memory
Finds the right fact, sees how records connect and remembers context.
Agent harness, tools, model gateway
Plans each step, uses approved tools and picks the right model.
Guardrails, evals, traces and audit
Checks each step, tests quality and keeps a full record.
02
Customer configuration
Labels stay, values change
Domain ontology and data model
What your records mean
Claims, invoices, service records and part photos, linked.
Customer rules and exception logic
Your warranty policy and coverage rules.
Integration APIs and approval path
SAP in, SAP out. Your reviewers approve.
Dashboards, worklists and SLAs
Agreed response times
A risk worklist, riskiest claims first.
03
Production AI worker
Labels stay, values change
Evidence-backed recommendations
Pay, Refer or Reject, with the proof.
- Pay
- Refer
- Reject
Deterministic validation checks
Fixed checks that give the same answer every time
Reused photos, duplicate invoices and missing documents.
Human review and approval gates
Your warranty team makes the final decision.
Monitoring, feedback and rollback
Each flag shows its weight and its source.
In every solution above, the core column stays the same. Only your data, rules, links and controls change.
Seven layers, from your files to your people.
Data comes in at the bottom. People use it at the top. Every layer checks who is asking.
Sample question
Can we pay voucher SV-1042?
Your network
07
Where people use it: chat, search, work lists, Microsoft 365 add-ins and APIs.
Your own screens can call it too.
06
Plans, acts and checks each step. It picks a model by how sensitive the data is.
Engineers call this the agent harness.
05
Checks permissions first, finds the right facts and cites them. If the facts are not there, it says so.
Search that respects access rights at the source.
04
Remembers context for a person, a team or the company, with rules.
Kept per person, per team or for the company.
03
Stores what it reads, with versions and who may see what.
Search indexes and a knowledge graph.
02
Reads every file, even scans, and keeps it in sync.
Text and image reading (OCR and computer vision).
01
Where your data already lives: ERP, email, file shares and databases.
Engineers call these connectors.
Cloud model: only if you approve it, and only for data you allow.
26 tested modules, in six groups.
We call them accelerator modules: ready parts for common jobs. Each one is tested in real work. New AI workers reuse them, so they ship in weeks.
26 modules
- 5 modules
Read and ingest
- Any-format extraction
- Large-volume extraction and Q&A
- Voice transcription
- Voice translation
- Speaker identification
- 3 modules
Classify and organise
- Document classification and sorting
- Conversation to notes
- Document summaries
- 5 modules
Match and correlate
- N-way document matching, many documents at once
- Document similarity matching
- Image similarity matching
- Cross-modal correlation, linking text, images and numbers
- Entity knowledge graph, a map of how people, parts and records connect
- 5 modules
Reason and decide
- Rule and tolerance checks
- Anomaly and fraud detection
- Confidence scoring, how sure the AI is
- Report intelligence
- Spec and BOQ parsing, bills of quantities
- 5 modules
Ask and act
- Plain-English Q&A
- Private document chat
- Real-time lookup and eligibility
- System posting and actions
- Connector framework, new data sources added in days
- 3 modules
Govern and prove
- Source citation and audit trail
- Access and permission inheritance, people see only what they may see at the source
- Private deployment, on your servers or in your region
Every step is logged. You can replay it.
Here is one sample voucher check. Each step is saved with its source. Try the stop switch while it runs.
- Step 01
Read 6 documents
Queued
- Step 02
Sort by type
Queued
- Step 03
Match amounts and payee
Queued
- Step 04
Check the VAT rule
Queued
- Step 05
Unsure, ask a person
Queued
- Step 06
Release the payment
Queued
- Step 07
Write the audit log
Queued
Flip it while the run is going. The AI worker stops at once.
Your data, rules and logs stay with you.
It runs on your servers or in your region. Your data is never used to train models.
Security is built into every layer, not added at the end. Here are four things that could go wrong, and what stops each one.
What could go wrong
Your data leaves your network.
What stops it
Your servers or your region
Data stays on your servers or in your region.
Your keys
Encryption with keys you hold.
No training
Your data is never used to train models.
What could go wrong
The wrong person sees data.
What stops it
Company sign-on
Single sign-on with your company login. Every request carries the person who asked.
Roles and teams
Access by role and by team.
Leak checks
Leak checks (DLP) and retention rules.
What could go wrong
An AI worker takes a wrong step.
What stops it
Tests before launch
Each AI worker is tested before it goes live.
People approve
People review risky steps before they happen.
Stop switch
A person can stop a run at once.
What could go wrong
Nobody can explain a result.
What stops it
Full trail
A full trail from each answer back to its source document.
Replay
Every step is logged, so you can replay a run.
Certifications and compliance
- ISO 27001 certified
- SOC 2 Type I report (available on request)
- GDPR compliant
- HIPAA compliant
Built for large companies. Ready for review.
Our AI Platform runs real work every day. It grows with you, fails safely and comes with the papers reviewers ask for.
It grows with you.
Search indexes are split by team, so answers stay fast as your files grow. Small models do the routine work, so costs stay low. Each new AI worker reuses the core.
It fails safely.
Every failure has a planned answer. If a model is down, it still finds and cites the facts. If a source system is down, it shows the last copy and says so. A bad update is stopped and rolled back.
It passes review.
Every change is tested, approved by named owners and can be rolled back. Your risk team gets the papers it needs.
What reviewers get
A fixed licence, not a bill per question.
Small models do the routine work where they fit, so running costs stay low. You choose where it runs.
Three ways to run it
On your servers.
Nothing leaves your network.
In your region.
Data stays in your region. You hold the keys.
With an approved cloud model.
Used only if you approve it, and only for the data you allow.
| Question | Typical cloud AI | Our AI Platform |
|---|---|---|
| Where it runs | Usually the vendor's cloud | Your servers or your region |
| How you pay | Usually per token or per seat, rising with use | A fixed licence |
| Who keeps the work | Usually set by the vendor | You keep your data, reports and roadmap |
| How answers are checked | Varies by tool | Every answer cites its source, and risky steps wait for a person |
Where it runs
Typical cloud AI: Usually the vendor's cloud
Our AI Platform: Your servers or your region
How you pay
Typical cloud AI: Usually per token or per seat, rising with use
Our AI Platform: A fixed licence
Who keeps the work
Typical cloud AI: Usually set by the vendor
Our AI Platform: You keep your data, reports and roadmap
How answers are checked
Typical cloud AI: Varies by tool
Our AI Platform: Every answer cites its source, and risky steps wait for a person
See the detailed comparison
| Compared on | Our AI Platform, built with you | Build it yourself | Cloud AI services you assemble | Chat assistant | Consultancy-led programme |
|---|---|---|---|---|---|
| Real strength | A tested core, plus your team's skills | Full control | Scale and a wide range of services | Fast to switch on | Strategy and change support |
| Where it starts | A core already in production, set up for you | A blank page | Parts you put together and run | A fixed product | Discovery, then a build on a stack they choose |
| Where your data runs | Your servers or your region, with your keys | Wherever you build it | Often their cloud and their models | The vendor's cloud | Depends on the stack |
| Proof behind each answer | Every answer cites its source, or it is dropped | If time allows | Up to you | Links, rarely checked | Varies |
| People approve risky steps | Built in: a named person approves | Yours to build | Yours to build | Limited | Varies |
| Running cost | Small models and a fixed licence | Your team and your hardware | Often per token, rising with use | Per seat | Day rates, then the platform's costs |
| Who owns it after | Your team, with no lock-in | Your team | You, tied to their platform | The vendor | Often the consultancy |
Our AI Platform, built with you
- Real strength
- A tested core, plus your team's skills
- Where it starts
- A core already in production, set up for you
- Where your data runs
- Your servers or your region, with your keys
- Proof behind each answer
- Every answer cites its source, or it is dropped
- People approve risky steps
- Built in: a named person approves
- Running cost
- Small models and a fixed licence
- Who owns it after
- Your team, with no lock-in
Build it yourself
- Real strength
- Full control
- Where it starts
- A blank page
- Where your data runs
- Wherever you build it
- Proof behind each answer
- If time allows
- People approve risky steps
- Yours to build
- Running cost
- Your team and your hardware
- Who owns it after
- Your team
Cloud AI services you assemble
- Real strength
- Scale and a wide range of services
- Where it starts
- Parts you put together and run
- Where your data runs
- Often their cloud and their models
- Proof behind each answer
- Up to you
- People approve risky steps
- Yours to build
- Running cost
- Often per token, rising with use
- Who owns it after
- You, tied to their platform
Chat assistant
- Real strength
- Fast to switch on
- Where it starts
- A fixed product
- Where your data runs
- The vendor's cloud
- Proof behind each answer
- Links, rarely checked
- People approve risky steps
- Limited
- Running cost
- Per seat
- Who owns it after
- The vendor
Consultancy-led programme
- Real strength
- Strategy and change support
- Where it starts
- Discovery, then a build on a stack they choose
- Where your data runs
- Depends on the stack
- Proof behind each answer
- Varies
- People approve risky steps
- Varies
- Running cost
- Day rates, then the platform's costs
- Who owns it after
- Often the consultancy
Nine AI workers run on this core.
Each one does one job, from warranty claims to invoices and dental care.
Questions reviewers ask us.
What is the AI Platform?
The tested core under every AI worker we build. It has 7 layers and 26 tested modules, reused each time.
Does any data leave our network?
No, not when it runs on your servers. In your region, it stays in that region. A cloud model is used only if you approve it.
Is our data used to train models?
No. Your data is never used to train models.
Which models do you use?
Small and large models, picked by how sensitive the data is. You approve the list.
Does it work with our ERP?
Yes. It uses your existing APIs first, and our connectors where needed.
Can we see every step?
Yes. Every step is logged with its source, so you can replay any run.
What is MCP?
A standard way for AI to use your systems. It allows approved actions only.
How is it priced?
A fixed licence, not a bill for every question or token.
Can we leave?
Yes. You keep your data, reports and roadmap.
Try it on your own documents.
Start with the 2-week assessment. We bring our AI Platform to your servers.
It starts with a short call with our AI experts.