OpsLedgerGroup operations console Demo by TheGradleW

Live Working demo for a multi-business owner

Three businesses.
One AI operations hub.

An AI agent that answers from your accounts, CRM and invoices; an inbox that routes and drafts replies; supplier bills read straight into Xero format with every sum re-checked; and a management pack that writes itself. Built hands-on, one workflow at a time.

  • LiveAgent picks its own tools (financials, overdue invoices, pipeline, forecast) and shows every call it made
  • LiveInbox triage: routes to the right business, extracts details, proposes approval-gated actions
  • LiveInvoice extraction with code-level checks: catches a supplier's wrong total
  • LiveManagement report where every £ and % is checked against the books

◆AI runs server-side on Amazon Bedrock (Kimi K2.5, Nova Pro fallback) with no keys in the browser. The model is one config value: switching to Claude via the Claude API or Bedrock needs no code change.

Group today

Sample data
12-mo revenue–
12-mo profit–
Overdue–

01 · Business dashboard Sample data

Every business on one screen

Revenue, costs, margin, receivables and pipeline per business or for the whole group, with a 3-month forecast (trend × seasonality, 80% range). In production this reads Xero, the CRM and booking systems on a schedule.

Revenue, costs & forecast

RevenueExpensesForecast

Overdue invoices

InvoiceClientAmountDays

Sales pipeline

From the CRM

Connected systems

Connectors are simulated in this demo. Each one is a standard API (Xero, Microsoft Graph, HubSpot, Stripe, GoCardless, Google) wired through n8n/Make or a small service, whichever is cheaper to run and easier for you to own.

02 · AI agent Live

Ask about any business in plain English

A real tool-using agent: the model decides which data tools to call, the server runs them against the same numbers you see above, and the answer can only use what the tools returned. Every call is shown.

Ask a question

Tools this agent can call

get_financialsmonthly P&L per business or group
get_overdue_invoicesunpaid invoices past due
get_pipelineCRM leads by stage
get_forecastnext 1–6 months with range
find_anomaliesmonths with unusual margins
list_businessescompanies and their software

Agent run

Pick a question or type your ownThe agent's tool calls and its answer appear here.

03 · Inbox automation Live AI Simulated systems

One shared inbox, routed and drafted

Each email is matched to the right business, contact details are pulled out (and dropped if they are not really in the email), next steps are proposed from a fixed list of allowed actions, and a reply is drafted for a human to approve.

Inbox · 4 new

AI triage

Select an email and press "Triage with AI"Routing, extracted details, proposed actions and a draft reply appear here.

Activity log

Simulated writes to CRM, booking & task systems
WaitingApprove an action above and it is recorded here.

04 · Document processing Live AI

Supplier bills in, Xero-ready bills out

The AI reads the invoice and suggests a business and account code. Then plain code re-checks every line, the VAT and the total, so a wrong supplier total is caught before it reaches the books. Try the second sample.

Production version takes PDFs and photos from an inbox or shared folder (OCR first, then the same steps).

Draft bill

Pick a sample or paste an invoiceFields, line items and the arithmetic checks appear here.

05 · Management reporting Live AI

The monthly pack writes itself, and checks itself

Numbers are computed in code from the books. The AI only writes the commentary, and every £ amount and percentage in its text is matched back to the figures. Anything it calculated on its own is highlighted for review.

In production this runs every Monday at 07:00, updates a Google Sheet / Excel workbook and emails a PDF.

Choose a business and monthTry Brightline Cleaning · Jun 2026 to see a cost spike explained.

06 · Workflow automations Simulated run

Automations you can see, test and own

The live features above, wired into scheduled and event-driven workflows. Each is built on whatever fits: n8n (self-hosted, cheap at volume), Make or Zapier (quick for the team to edit), or a small coded service when logic is heavy. Press "Test run" to step through one.

07 · How I'd run it

Managing technology across several businesses

One owner, many ideas, several companies. The trick is a shared foundation, a single prioritised backlog, and short weekly cycles so something useful ships every week.

1

Audit & map (week 1)

List every system, login, data flow and manual task per business. Score each idea by hours saved, money at stake and effort.

2

Shared foundation

One integration layer (n8n or Make + a small API), one place for credentials, logging and alerts, reused by every business.

3

One backlog, weekly ships

A single board across businesses. Each week: one automation or feature live, measured, and documented in plain English.

4

Own it, maintain it

Monitoring and failure alerts on every workflow, monthly cost review, and handover notes so nothing depends on one person.

What is real in this demo

PartStatus
AI agent with tool callsLive real model, real tool execution on sample data
Inbox triage & reply draftsLive any email you paste
Invoice extraction & checksLive any invoice text you paste
Management report + number checkLive
Dashboard, forecast, CSV exportLive computed in your browser
Businesses, invoices, CRM dataSample fictional, generated
Xero / HubSpot / M365 writes, workflow runsSimulated no external systems touched

Stack used here & proposed

Claude API / Claude CodeAmazon BedrockTool use / function callingAWS LambdaCloudFront + S3JavaScript / TypeScriptNode.jsReactPythonn8nMakeZapierXero APIMicrosoft Graph (M365)HubSpotGoogle Sheets APIStripe / GoCardlessWebhooksPostgreSQL

Built and hosted by TheGradleW: static site on CloudFront, one Lambda for the AI endpoints, no API keys in the browser, security headers on every response.