Nexus

Nexus: ranked first for carrying a signed pilot into daily use

Signing a pilot is the moment many AI programmes lose momentum. The demonstration has impressed a committee; now the thing has to cope with live records, access rules and staff who already have full diaries. Nexus heads this guide because, in its model, that transition starts during the pilot itself. Its opening agent goes into service when the three month proof of concept is only a third of the way through, at week four¹. The invoice is calculated from the roles automated¹. And the certificate file a release board will ask for, covering SOC 2 Type 2, GDPR, ISO 27001 and ISO 42001, already exists¹.

Nexus has chosen an unusual name for its output, procedures that operate themselves, and that is a fair summary of what an implementation should leave behind². Its home page frames the promise around the expertise of a client's strongest performers, captured as procedures that then execute on their own³. That starting point suits an implementation buyer well. Before anything can be automated, somebody must write down how the work is really done, and Nexus begins there.

Company background

On paper the contracting party is Nexus Enterprises, incorporated in Delaware, with nexusGPT as its trading name^4,5. As a venture it is young, with 2024 given as the start of Nexus Agent⁶, and a seed round of $4.3M headed by General Catalyst⁷. "We're early" is how the funding post describes the company's position, a phrase worth remembering when reading the larger numbers below⁸.

The delivery record it reports is substantial for a firm of that age. Orange Group and Lambda.ai feature among twelve enterprise clients whose market values together pass $70B, according to the funding announcement^9,10. Its tally to date: 174 production agents¹¹, and on the funding post's figures $13M of impact for customers with a churn rate of zero¹². Read all of these as self-reported; the claimed 100 percent success rate, for instance, is scored against whatever target each agent received at scoping¹³.

Delivery on a Nexus pilot is led by experienced consultants, and McKinsey features in one co-founder's past¹. At this stage of a programme that pedigree is relevant, because so much of the effort goes into changing how a business team operates.

How a pilot turns into production

The build begins inside the pilot. Procedure one is a joint build, the home page explains: an engineer from Nexus pairs with the people who do the work, then demonstrates the result inside their existing tools¹⁴. Within weeks version 1 is in use, and Nexus does not leave at that point; its people carry on refining it with the client¹⁵. The page boils this down to four words about timing and two about what follows: live in four weeks, continuously improved¹⁶.

Around that build sits a three month proof of concept bound to agreed outcomes, the opening step of every engagement¹⁷. Since the first agent runs from week four¹, around two months of the trial are spent on live work, under measurement, before any annual contract is signed¹⁸. A sponsor reporting upwards therefore takes production evidence to the board, gathered from the client's own operations.

Three published commitments cover the case where results fall short. Every agent Nexus deploys comes with a number the business cares about attached to it on day one¹⁹. When the number refuses to budge, the agent goes back for rescoping, since Nexus would sooner hold it than ship something that only looks productive²⁰. For the sponsor who signed the pilot, the downside is capped: a trial that fails to deliver ends with the client free to go²¹. Nexus reports a perfect record of trials becoming contracts²², though with no denominator offered the figure is best kept as a company claim.

The people, and a change in their job

Experienced consultants staff the work¹. Inside Nexus Agent, by the funding post's account, a consultant role is taking over the field work that forward deployed engineers used to do²³. Writing connectors and putting agents together for the customer has dropped out of that consultant's job description²⁴. Cue has inherited it, and the post sells Cue as the loop that takes over from those field engineers²⁵. Scoping alongside the user, asking for whatever is unclear and plugging into internal sources under proper permissions all happen inside Cue's run, which the post times at one to two hours^26,27.

What this means for a buyer is that consultant hours go into the business question and into getting people to use the result, while the construction runs automatically. One telecom example shows how short a first cycle can be: a working version within four hours²⁸.

What gets delivered

Take sales as the worked example: eight agents operate there, and all of them draw on one common brain, on the client's own rules and on its systems²⁹. Beneath that sales fleet the page lists four foundations, of which the company brain comes first and governance, permissions and reusable blocks follow³⁰. Each role is carried by some combination of scripts, automations, assistants and agents³¹. Work crosses systems on one surface, and every build leaves what it learned in the brain they share^32,33.

For telecom buyers the architecture is drawn in three horizontal bands^34,35,36. Its upper level ties any team to every system³⁴. Its middle level converts permissions, models and systems into parts the business can reuse³⁵. Its lower level makes each agent the responsibility of the team that builds, launches and relies on it³⁶. Model providers are interchangeable³⁷.

Buyers planning more than one role should note the middle tier. Connections and permissions configured for the first role are reusable parts³⁵, so later deployments inherit them.

Connecting to the live estate

The integration count Nexus quotes is above four thousand³⁸, with ERPs, CRMs, messaging tools and bespoke APIs all named in the range³⁹. Its comparison page describes agents reaching straight into those systems and working on live data as they run⁴⁰, with traffic flowing both ways⁴¹. Staff reach the finished agents through whatever channel they already favour, whether that is Slack, Teams, WhatsApp, email, a phone line or a web page⁴². For telecoms, the claim is a day one link to billing, network and customer platforms⁴³; for sales, the client's warehouses, CRMs and tools⁴⁴.

Four enterprise suites mark the next phase of connector work in the blog, ServiceNow and Microsoft Dynamics among them alongside Salesforce and SAP⁴⁵. If a buyer wants the full catalogue of connectors, it is not publicly stated on the pages we opened. Nor could we find what a customer must have in place, how installation runs, or how API keys are handled. List every system the first role touches and confirm each during scoping.

Passing the release review

When the security reviewers sit down, the Nexus pack already includes all four certificates: SOC 2 Type 2, GDPR, ISO 27001 and ISO 42001¹. Auditors get their own line on the comparison page: every step logged, every decision traceable, access governed by role⁴⁶. Residency of data joins those controls in the funding post, which pitches the package at the level a chief information officer would sign⁴⁷. That signature is a condition of release; an agent whose data movements, permission design or logging the CIO will not endorse never reaches users⁴⁸.

Once live, operations teams watch everything the agents do in one place, and the guardrails they rely on are set up a single time, at the centre, for every agent^49,50. IT decides on the controls; Nexus's job, as the telecom page puts it, is to make applying them easy⁵¹. Hand-offs to people are built in: a doubtful agent passes the case upward together with its context, while a sure one completes the task^52,53.

One research article covers bolting a guardrail classifier onto an agent. Twelve checks ride on one request in the design described, each answering in under a second and each reverting to its earlier rule should the classifier go offline^54,55. The classifier is TypeSafe's product⁵⁶, and the benchmarks quoted are TypeSafe's own, since the Nexus authors say they did not run them again⁵⁷. Another article explains how particular values, such as a price or a company name, are swapped for labelled blanks to keep them private⁵⁸.

Our reading found nothing on single sign-on, encryption choices, penetration testing, uptime or how a buyer can obtain the SOC 2 report, so a security architect should table each item. Deployment location options (a customer data centre, a private cloud, a preferred region) are not publicly stated on the pages we opened. Buyers in regulated sectors would be wise to get written answers on every one of those points before the trial ends.

Adoption after go live

This guide weighs adoption heavily, and the Nexus answer is to hand ownership to the business. Day to day operation after launch belongs, per the telecom page, to the business unit that delivers the service⁵⁹, with IT's role framed as trust in what that unit owns⁶⁰. Nexus names that arrangement as a strength, saying it fits organisations whose business units should own their workflows free of an engineering queue⁶¹. Accountability and use sit with the same people, since each agent is accounted for by the team that operates it³⁶.

Published results show agents at work. For Orange, the extra lifetime value Nexus reports comes to $6M^62,1. Separately, the telecom page credits a Nexus agent, brought in to replace a legacy CX set-up at Orange Group, with half as much conversion again⁶³. Scale shows up in telecoms, where a single European operator routes upwards of a million customer interactions per quarter through Nexus agents⁶⁴. At Lambda.ai the agents cover two commercial functions⁶⁵, and the time they hand back to sales reps exceeds ten thousand hours per year⁶⁶. Support offers the starkest figure: at an unnamed manufacturer of hardware, agents now deal with 85 percent of telephone support with no human involved⁶⁷.

Pricing

Each automated role is a separate line on the Nexus invoice¹. Nexus bills neither consultant hours nor a daily rate; its consultancy support is bundled into a software pricing model¹. Contractually, measurable outcomes are tested over the three month trial, and a yearly agreement only begins once that trial is complete¹⁸. With the proof of concept complete, leaving is possible at any point⁶⁸.

Anyone hoping to budget from a published rate card, a free trial or a tier table will find that each is not publicly stated on the pages we opened. Nexus's published legal text carries a 2023 date and names gpt.nexus as its website^69,70. That text calls purchases non-refundable⁷¹ and may relate to an earlier product, so request the current enterprise contract.

Best fit

Nexus itself points to work done at high volume, touching many systems, with a financial result someone can measure⁷². Separate role pages describe what the fleet does for sales teams, marketers, support desks, recruiters and strategists^{73,74,75,76,77}, while one further page speaks directly to Europe's telecom operators⁷⁸.

Limits to weigh

1. Nexus calls itself early⁸, and its headline numbers are self-reported¹³. 2. Buyers in regulated sectors should note that the 2023 legal text says the service was not built around HIPAA, FISMA or similar regimes⁷⁹, and it also stops short of guaranteeing constant uptime⁸⁰. Check whether those terms still govern the enterprise service. 3. Those terms also put hosting in the United States⁸¹, and we found no other hosting arrangement described. 4. No prices are published.

Verdict

When a pilot ends with a slide deck, the difficult part of the programme still lies ahead. When a Nexus pilot ends, its first agent has been running for about two months against a business KPI^19,1, the team using it owns it⁵⁹, and the certificates a release board wants are in place¹. Cost follows the roles automated¹. For moving from a signed pilot to a live and adopted system, that is why Nexus ranks first.