Put AI to work. Then teach your team.

Self hosted AI on your hardware, built into the work you already do, with a person approving what matters. Then hands on training so your team uses it well.

Book a call

Part 1Build AI into the work Part 2Train your team to use it

Your work, your hardware

Invoice 2291 readSupplier, amount and due date pulled out

Routed to accountsClassified as a bill to pay

Waiting for a personNothing is paid without approval

Part 1Building AI in

AI does the busywork. People make the calls.

We start with one piece of work your team repeats every day, like reading invoices or sorting requests, and build AI into that. A person still approves anything that matters.

How one item moves throughRuns on your hardwareEvery step logged

Comes inEmailsScanned documentsWeb formsShared folders

OrchestratorLeads

Runs the steps in order and keeps a record of each one. It hands each step to the part built for it.

01Reads

Intake

Pulls the fields out of documents, emails and forms. Nothing gets typed twice.

Example output

SupplierExample Supply Co.

Amount$1,840.00

DueIn 30 days

02Decides

Classify and route

Works out what each item is and who it belongs to, then sends it to the right system, record or person.

Example output

TypeBill to pay

Goes toAccounts

RuleOver $1,000 needs a yes

03Does

Act

Files it, updates the record, and flags only what needs a human.

Example output

RecordBill drafted

FlagWaiting for a person

Log3 steps recorded

Goes out

Records updatedFiled in the system it belongs toA person decidesAnything unclear or over a limit waitsAn audit trailWhat it did, when and why

Try it. Example emailsPick an email. The AI reads it, then you decide.

What the AI read

Pick an email on the left.

What happens next

TypeWaiting

RouteWaiting

Audit trail

On your hardware

Self hosted. No subscription traps.

Your data, your choice

Self hosted, or outside models with guardrails.

A person approves

It flags what needs a human and waits.

Every step logged

You can see what it did and why.

Built with

n8n workflow automationDocument pipelinesYour existing systems

Your AI tools

Tie your tools together. One setup, your rules.

Your team may already use Claude, ChatGPT, Copilot or Gemini. We connect them, and your own self hosted models, to one setup. Every tool sees the same context and follows the same rules.

How the tools connectRuns on your hardwareOutside service, through your guardrails

The tools you use

ClaudeOutside model

ChatGPTOutside model

CopilotIn Microsoft 365

GeminiOutside model

OllamaRuns on your hardware

One setup on your network

RoutePicks the right tool for each jobMemoryKeeps the shared contextGuardChecks every request against your rules

What your team gets

Shared contextEvery tool works from the same files and history.

Task routingEach job goes to the tool that suits it.

Document actionsRead, file and update records.

GuardrailsYour rules for what may be sent where.

Audit trailEvery step logged, on your network.

One setup across every toolShared memory and task routingGuardrails and an audit trail on your network

Where to start

Three ways we put AI to work.

Most teams start with one and add the others later. They are built to work together.

Your server

ModelsOllama

Workflowsn8n

DocumentsPaperless-ngx

A self hosted stack

Your hardware. Your data.

We set up the whole stack on your infrastructure: a server in your office, a machine in your data center, or a box we ship and set up for you. Everything talks to each other and stays on your network.

Ollama runs the models, n8n runs the workflows, Paperless-ngx keeps the documents
We look after hosting, updates and uptime
One flat fee, no per user subscriptions
Your team owns the models and the data

Submittal packageScanned, read, indexed

SummaryOpen itemsDraft reply

Document processing

Your documents, read and filed.

Documents land in one place, get scanned, read and indexed, then a workflow sends them to your model. Summaries, open items and draft replies are ready when your team starts the day.

Submittal logs and tracking
RFI drafts and follow ups
Invoices routed to the right cost code
Contract clauses pulled out and flagged

AI tools check

What you pay for

What you use

Example

Training and adoption

Know what you are buying. Know how to use it.

A lot of what is sold as AI is the same few models with a new price. We check what you pay for against what you need, then train your team on the tools that fit, inside their actual work.

A check of the AI tools you pay for
Hands on workshops by role
Prompts written for your industry
Support as the tools change

See the training plan

A training room seen from the back of the room, with a slide titled Week 2: Writing with AI

Tools only help if people use them. So we teach them.

Part 2 is training, built around the tools you own and the work your team does every day.

Part 2: Training

Part 2Training your team

Training with your own work. Not generic slides.

Your team practices on real documents and real tasks, on the tools you own. Each week builds on the last, and we check where people are before moving on.

Team confidence, by weekWeek 1

Week 1This weekExample values

Hands writing notes next to a laptop and a glass of water

Week 1

The basics, on your documents

What the tools can and cannot do, tried on your own files.

Hands with a tablet and a coffee on a wooden table

Week 2

Writing and checking

Drafting emails and reports, and how to check the output before it goes anywhere.

A long empty conference room with rows of tables

Week 3

Your workflows

Using the systems from Part 1 day to day, and what to do when something is flagged.

Week 4

On their own, with a check in

People use it on their own work. We review together and fix the gaps.

How it works

Two tracks. One team running both.

You can do both parts, or just one. Most teams start with one workflow and train alongside it.

Part 1. Building it in

Pick one workflow

We find the repeat work that costs the most time and agree what the AI may and may not do.

Build it on your hardware

We set up the stack on your machines and connect it to your inbox, files and systems.

Run it with a person approving

It works alongside your team. Anything unclear is flagged for a person.

You getA working system and an audit trail

Part 2. Training

See where people are

A short check of what each person knows and what they need.

Weekly sessions on real work

Hands on, on the tools you own, with practice between sessions.

Check in and follow up

We review how it is going and help with the hard cases.

You getA team that uses it with confidence

Before you automate

AI is only as good as the data it can reach.

A lot of companies have years of documents in folders nobody has touched, and processes that live in one person's head. Before you automate, you organize. We get your systems ready so the AI works when it goes live.

Example. Made up folders.

Today

Scans 2019 FINAL (2)No clear owner

Invoices oldMixed years

Job files miscThree naming styles

Copy of price listWhich one is current?

How we quoteOnly in one person's head

Ready for AI

Jobs

Job 14

InvoicesNamed by date and supplierDrawingsCurrent version onlyPhotosTagged by job

Price listsOne current list, old ones archivedHow we quoteWritten down step by step

Document structure audit

We map what you have, where it lives and what state it is in.

Migration plan

A step by step plan to move old data into systems your AI can use.

Ongoing data hygiene

Processes and automation that keep your data clean as the business grows.

Example scenario. Not a real client.

An office buried in invoice emails.

How a small office might use both parts. Names and details are made up.

1Before

One person read every supplier email, typed invoices into accounting and chased approvals by hand.

2What we did

Part 1: an intake workflow that reads invoices and routes them for approval. Part 2: four weeks of training for the office team.

After

Invoices arrive filled in and waiting for a yes or no. The team knows how to check the work and when to step in.

FAQ

Questions, answered.

The short version: AI on your hardware, a person approving, and a team trained to use it.

Still wondering?A real person reads every message.

Email us

Not unless you choose it. By default we run the AI on your own hardware, so documents and emails stay on your network. If your team wants outside models like Claude or ChatGPT, we connect them through guardrails you set, and every request is logged.

Yes. We connect the tools your team already uses to one setup, with shared context, rules for what each tool may see, and a log of what was sent.

No subscription traps. The stack runs on hardware you own. We will tell you up front if your setup needs anything new.

Only the ones you allow. We agree the rules with you, and anything outside them is flagged for a person.

Often not yet, and that is normal. We start with a document structure audit, then plan the move into systems your AI can use.

Yes. We can train your team on the tools you already have.

Weekly hands on sessions in small groups, using your own documents and tasks, with a follow up after the last session.