Find your biggest AI opportunities in under 30 minutes.Book a consultation
Back[ Coventa ]Case Study

[ AI-NATIVE BUSINESS PLATFORM ]LIVE

Nexus

The business platform that runs itself. Built on a mature open-source ERP core and rebuilt as AI-native across CRM, ATS, ERP, HRM, project and time management, invoicing and inventory — where agents don't just record work across modules, they do it.

NEXUS // UNIFIED WORKSPACE

// Problem

The Problem

The mid-market back office is assembled from six tools that each own a fragment of the same entity. A candidate in the ATS becomes an employee in the HRM, works on a project in the PM tool, logs time in a fourth system, and appears on an invoice generated by a fifth — with a person reconciling the seams. AI features bolted onto each tool individually make this worse: six assistants, each blind to the other five, each summarising its own fragment.

  • The same entity exists six times with six identifiers and no shared truth.
  • Cross-module work — hire to staffed to billed — is manual reconciliation, every cycle.
  • Per-tool AI assistants cannot reason across the boundary where the actual work happens.
  • Integration cost recurs at every vendor upgrade.

// Overview

Nexus takes a mature open-source business platform spanning CRM, ATS, ERP, HRM, project and task management, sales, financial management with invoicing and payments, and inventory — and rebuilds it around a unified data model with agents operating natively across modules. Because a candidate, employee, project allocation, timesheet and invoice line are facets of one model rather than five records, an agent can act along the whole chain: staff a project from available capacity, raise the timesheet, and generate the invoice, as one operation under approval. The open core keeps the platform inspectable and avoids vendor lock-in on the substrate.

// AI System

Why AI

Agents are only as useful as the boundary they can reason across. Bolting an assistant onto a single module reproduces exactly the silo it sits in — which is why per-tool AI features have delivered so little. The design decision here is architectural rather than model-driven: unify the data model first, then let agents operate over the whole of it. That is what makes “hire to staffed to billed” a single agent-executable operation instead of a process diagram.

// Specs

Specifications

MODULES
CRM, ATS, ERP, HRM, project & task, sales, finance, inventory
FINANCE
Invoicing, estimates, billing, payments, cost management
WORKFORCE
Time and activity tracking, allocation, capacity
FOUNDATION
Mature open-source core, extended AI-native
AGENTS
Operate across modules on a unified data model

// Features

Features

  1. 01One data model across CRM, ATS, ERP, HRM, projects, finance and inventory.
  2. 02Agents execute along the full chain — hire to staffed to timesheet to invoice — under approval.
  3. 03Time and activity tracking feeds capacity planning and billing from the same record.
  4. 04Invoicing, estimates and payments generated from project reality rather than re-keyed.
  5. 05Open-source core keeps the substrate inspectable and avoids lock-in.
  6. 06Extended AI-native rather than assistant-bolted, so reasoning crosses module boundaries.

// Architecture

Architecture

PLATFORM TOPOLOGY

Runtime · one item, left to right


  1. 01User / Agent Intent
  2. 02Unified Data Model
  3. 03Cross-Module PlannerCRMATSHRMProjects & TimeFinanceInventory
  4. 04Transaction Execution
  5. 05Approval Gate
  6. 06Ledger + Records

dashed = the inference step, where the system exercises judgment

System stack

Data in · decisions out

01

Sources

Six tools, one business

CRMATS & HRMProjects & timeFinance & invoicingInventory

02

Ingestion

Native modules, no reconciliation

Unified data modelMigration connectorsfrom incumbent toolsEvent busPermissions model

03

Ontology

One entity, many facets

Account · ContactProject · TimesheetInvoice · PaymentEmployee · CandidateItem · Stock

04AI

Intelligence

A planner across every facet

Plannerreasons across modulesSignal detectionrisk, churn, cashDrafting & automationForecastingEval suite

05Human

Human control

Owners act, approvers approve

Owner acts on recommendationApproval thresholdsRole permissionsAudit trail

06

Actions

Written back

All Nexus modulesEmail & calendarReports

Observability

Every model call traced; evals run on real cases, not anecdotes.

Governance

Entitlements enforced at retrieval; rules versioned by the organisation.

Write-back

Systems of record are written only through the approval gate.

One entity, many facets — the planner reasons across all of them rather than within one.

// Impact

Impact

One
Platform, replacing sixdesign intent

Interested in Nexus?

Let's map your current back-office stack against one platform.

Get in touch
// End of case studyNexus