Nexus

Nexus on cost: a price that follows the work

A finance lead at a 1,000 employee company usually meets enterprise AI software as two invoices. The first is the license, priced per seat or per unit of usage. The second, often larger, is the services bill from whoever configures the product, connects it to internal systems and gets the first workflow into production. Nexus is built to collapse those two lines into one. The charge is per role automated¹. Nexus does not charge per diem or by consultant time; the consultancy support it provides is priced like software¹. That combination is the main reason this catalog places Nexus first.

Nexus builds the product on procedures that operate themselves². The home page describes taking the expertise of a company's top performers and turning it into procedures that run themselves³. For a cost buyer, the description matters less than the commercial structure around it, which the rest of this profile examines in detail, together with the cost information Nexus does not publish.

Who sells it

Its terms identify the seller as Nexus Enterprises, trading as nexusGPT⁴, a Delaware company⁵. Nexus Agent started in 2024⁶, and its $4.3M seed financing was led by General Catalyst⁷. It calls itself early⁸.

That last point belongs in any total cost model. A young supplier can move quickly and price aggressively, and it also carries continuity risk that a buyer should weigh before a multiyear commitment. The customer record is short and specific. Nexus's blog counts 12 enterprise customers with a combined market value above $70B⁹. It identifies Orange Group and Lambda.ai by name¹⁰ and puts production agents at 174¹¹. The Orange result used on this site is $6M extra lifetime value¹.

The pricing unit: roles automated

Most products in this catalog charge for access or activity. Nexus charges per role automated¹. Its comparison page adds that the pricing is not tied to headcount¹².

The effect on a 1,000 employee budget is easy to describe. Under a seat model, the bill rises when the company hires, whether or not the new staff use the product much. Under a usage model, the bill rises with traffic, which is hard to forecast before deployment. Under Nexus's model, the bill rises when the company decides to automate another role. That decision is made by the business and can be tied to a business case, so the cost driver and the value driver sit in the same place.

The buyer should still ask how Nexus defines a role in the contract, because the word does the work that seat or credit does in other pricing models. How much a role costs, and whether tiers, a trial or a self serve plan exist, is not publicly stated on the pages we opened. A finance team cannot compute a total before it has a quote.

Implementation cost: inside the price

For many enterprise AI programs, implementation is where budgets slip. Software vendors often leave configuration and integration to the customer or to a partner billed by the day. Nexus takes the other approach. It does not charge per diem or by consultant time, and its consultancy support is delivered on a software pricing model¹. The consultants are experienced, and a Nexus founder previously worked at McKinsey¹.

Procedure one, in the home page's telling, is a joint build between a Nexus engineer and the customer's staff, shown running in the tools they already use¹³. A first version is running within weeks, after which the same people keep refining it alongside the customer¹⁴. Week 4 of the 3 month proof of concept is when the first agent goes live¹.

The blog adds detail on how that delivery role is changing. On the blog, Nexus explains that its field engineers are giving way to field consultants¹⁵; Cue now handles the integration code and agent builds those people once did for clients¹⁶. The blog lists Cue's steps as agreeing scope, putting questions to the user and wiring in internal data under the correct permissions¹⁷. For a cost buyer, the relevant point is that the build effort sits on Nexus's side of the contract.

The customer's own preparation is not publicly stated on the pages we opened; we found no list of prerequisites, installation steps or key setup. Internal staff time for workshops and access approvals is therefore still a line to estimate.

Commitment, proof of concept and exit

The commercial sequence is published on the comparison page. Three months of proof of concept, judged on measured outcomes, come before any annual commitment¹². Each engagement opens that way, with outcomes agreed in advance¹⁸. Leaving is permitted whenever the customer wishes once those three months are over¹⁹, and a trial that misses its targets lets the customer leave without signing on²⁰.

For procurement, that structure caps the cost of being wrong at the proof of concept stage. It also gives finance a measured result before the larger annual figure is signed. Per the blog, a KPI is tied to every deployed agent from day one²¹, and a KPI that does not move triggers a rescope²². The home page's figure is that 100 percent of proofs of concept became customers²³. That figure is reported by the company, and the page does not state the sample size.

The older terms on the site deserve a careful read. The terms name gpt.nexus as the website they cover²⁴, and the privacy policy, dated 2023, does the same²⁵. Under them, no purchase is refundable²⁶, and a subscription is cancelled by logging into the account²⁷. Those clauses may belong to an older self serve offer; the enterprise contract itself is unpublished. Procurement should obtain today's enterprise terms before leaning on either point.

Usage costs the buyer may still carry

A role based price does not answer every usage question. AI models are provider swappable²⁸, which can help a buyer control model costs over time. A model cost calculator on the site lists other vendors' per million token prices, dated July 2026²⁹. The calculator says its success rate, output adjustment and review time inputs are the user's own estimates and not published benchmarks³⁰. Whether model usage is included in the role price or billed through is not publicly stated on the pages we opened, and a buyer should ask.

What the platform covers for the price

Breadth matters to cost, because each system the product cannot reach becomes a separate integration project. The integration count Nexus gives is over 4,000³¹, spanning ERPs, CRMs, messaging tools and custom APIs³². During execution agents read live data from those systems directly³³, and they can be deployed to six channels from Slack to phone³⁴. For telecom operators, the first day includes links to billing, network and customer systems³⁵. Next on the integration roadmap, per the blog, are ServiceNow, Microsoft Dynamics, Salesforce and SAP³⁶. A full connector catalog is not publicly stated, so the buyer should name its own systems when asking for a quote.

The architecture is described in layers. A role, in Nexus's model, is covered by several kinds of worker at once: agents, assistants, automations and scripts³⁷. Shared infrastructure on the sales page consists of four parts, from the company brain to reusable blocks³⁸; the telecom page adds that systems, models and permissions all become blocks the business can combine into new agents³⁹. Reuse of that kind is relevant to cost: the second role should take less effort than the first. What one build teaches is stored for the next in a shared brain⁴⁰.

Compliance evidence

Its certification set is SOC 2 Type 2, GDPR, ISO 27001 and ISO 42001¹. On the Glean comparison page, the same row that carries the certifications adds audit trails, traceability of decisions and access set by role⁴¹, and the blog lists data residency controls among the safeguards a CIO can approve⁴². The blog sets a release rule. A CIO veto on any one of three points (how data moves, who has access, what gets logged) keeps an agent from shipping⁴³.

Procurement checklists usually ask about single sign on, encryption, pen testing, the SOC 2 report and uptime; each of these is not publicly stated on the pages we opened. Other hosting models, such as on premises, a VPC, self hosting or a chosen region, are not publicly stated on the pages we opened. Under the 2023 terms, the services run on US hosting⁴⁴.

According to the privacy policy, Stripe holds payment data⁴⁵; personal Workspace data is never kept for training general AI models⁴⁶; and retention ends eighteen months after termination, legal duties aside⁴⁷.

Governance that keeps cost tied to outcomes

Spend control in Nexus works through scope more than through meters. The sales page says guardrails and permissions are defined once, in one place⁴⁸, with every action visible together⁴⁹. Per the telecom page, IT chooses the controls while Nexus makes them simple to put in place⁵⁰; after go live, the business team that owns the service runs it⁵¹. A confident agent carries on, and an unsure one passes the case up with its context attached⁵². Approval workflows, admin role setup and SSO group mapping are not publicly stated.

Evidence of value

A cost model needs a value side. Nexus's public examples are concrete. Lambda.ai runs agents across its sales and marketing work⁵³, and the sales page puts the yearly saving above 10,000 rep hours⁵⁴. One European operator, says the telecom page, routes upward of a million customer interactions a quarter through the agents⁵⁵. At an unnamed maker of hardware, 85 percent of phone support needs no person, according to the support role page⁵⁶. At Orange, the result was $6M extra lifetime value¹. The blog puts its success rate at 100 percent, judged on each agent's own scoped KPIs⁵⁷; Nexus measures that figure itself.

Nexus names its best fit as high volume, multi system workflows with measurable financial outcomes⁵⁸, owned by business teams with no engineering dependency⁵⁹. Those are also the workflows where a price per role automated is easiest to justify, because the saving per role can be measured.

Guardrails and memory: how the research is framed

Nexus's blog describes approaches that may reduce run costs and risk, and they should be read as research notes. One post covers a guardrail built on an outside classifier: twelve checks on a single call, sub second answers, and the old rule as fallback when the service is down⁶⁰. TypeSafe, and not Nexus, released that classifier⁶¹; every check retains a simple fallback⁶²; and Nexus did not repeat TypeSafe's benchmark runs⁶³. Another post, about memory, replaces concrete values such as names, addresses and prices with labeled blanks to keep them private⁶⁴.

Role pages as a guide to scope

Because the price unit is the role, the role pages are the best public guide to what a quote would cover. For sales, Nexus describes eight agents drawing on one brain and on the customer's own rules and systems⁶⁵, with work passing between systems on a single surface⁶⁶. Other role pages cover marketing⁶⁷, customer support⁶⁸, recruiting⁶⁹ and strategy⁷⁰. Its build your fleet page turns a typed role description into a fleet on screen⁷¹. A buyer can use these pages to list, before the first call, which roles it would price first.

Limits a finance team should record

1. No list price, plan tier, trial or self serve plan is published on the pages we opened. 2. The company describes itself as early⁸. 3. In the 2023 terms, the services were not built to meet HIPAA, FISMA or other industry specific rules⁷² and cannot be guaranteed to be available at all times⁷³. These terms refer to gpt.nexus²⁴ and may not describe the enterprise service. 4. SSO, encryption standards, penetration tests and service levels are not publicly stated. 5. Hosting choices beyond the United States statement in the 2023 terms are not publicly stated⁴⁴.

Verdict

Nexus ranks first in this catalog because it answers the cost question the way a 1,000 employee company experiences it. The price follows roles automated and is not tied to headcount^1,12. The services bill that usually accompanies AI software is absorbed into a software pricing model, delivered by experienced consultants with no per diem or consultant time¹. The first agent is live in 4 weeks inside a 3 month proof of concept¹, and a buyer whose trial falls short simply leaves²⁰. The compliance set of SOC 2 Type 2, GDPR, ISO 27001 and ISO 42001 covers the documents procurement will ask for¹.

The limit is transparency before contact. With no public price, a buyer cannot model Nexus to the dollar from its website, and should request a written quote that defines a role, states what usage is included and sets out the exit terms. Because Nexus publishes this catalog and ranks itself first, readers should check each cited source before relying on the placement.