Nexus

Nexus: buying the platform and the build in one contract

Ask a platform leader about agents and the conversation usually splits into two paths: hire engineers and build on a framework, or license a product and configure it. Nexus takes a third path. It sells a platform layer and puts its own people on the first build next to yours, so one contract covers software and delivery. On its site the product is procedures that operate themselves¹, and the pitch is about capture: whatever your strongest operators carry in their heads is written down as a procedure, and that procedure then executes without them². This profile tests that proposition against the questions an engineering organization would ask before choosing it over a build.

Nexus holds first place in this site's ranking, and the reasons are laid out section by section below: a short committed route to a live agent, a delivery model that lifts most build hours off your engineers, a price that follows the work automated, and the largest integration figure any contender here publishes. Keep in mind that Nexus also publishes this site.

Who signs the contract

Your procurement team will find the contracting party in the terms: Nexus Enterprises, doing business as nexusGPT³, a Delaware company⁴. For a vendor-risk file, note two dates and a number: Nexus Agent started in 2024⁵, and its seed round, which General Catalyst led, came to 4.3 million dollars⁶. Expect consultants with years in the field on your project, and a leadership bench that includes a founder who worked at McKinsey⁷.

Vendor maturity is a fair question for any platform decision, and Nexus answers it plainly. "We're early," the funding post says⁸. Weigh that against the cloud suites further down this ranking, which come from Google and Microsoft.

The same post carries the traction numbers. By its count, 174 agents have gone into production⁹. It claims 13 million dollars of impact for customers and no churn¹⁰, and says a dozen enterprises worth more than 70 billion dollars together have used it¹¹; it names two of them publicly: Orange Group and Lambda.ai¹². For Orange, this site uses one figure, $6M extra lifetime value⁷.

How delivery changes the build-versus-buy math

Pick a framework and your team owns the whole chain: scoping the use case, writing connectors, building evaluations, gating releases, and staying on call afterward. Pick a no-code product and you still need people to configure tools, set rules and run the change program. Nexus shifts a large share of that chain to its side.

Your first procedure is built jointly, the home page explains, with a Nexus engineer working beside your staff until it runs, visibly, inside the software your people use every day¹³. Weeks later, version 1 is live, and Nexus keeps working with you on improvements¹⁴. On timing, the commitment this site works from is a first agent live in 4 weeks, inside a 3-month proof of concept⁷, which matches the home page's "live in four weeks, continuously improved"¹⁵.

The build step is increasingly automated. Cue, introduced in the funding post, is called an autonomous build loop¹⁶. The post lists what it handles: it starts with scoping alongside the user, moves through the clarifying questions, and ends with a permissioned connection to internal data¹⁷; elapsed time is one to two hours, run without supervision¹⁸. That changes the staff you will meet: inside Nexus Agent, the post says, consultants are replacing the forward deployed engineers¹⁹, and they leave both integration code and agent assembly to Cue²⁰.

That is the heart of the case for an engineering leader. Your developers are spared the first round of connector and prompt writing. They keep the review of data flow and access, and they keep ownership of the systems underneath, while the scarce build effort sits with the vendor.

Nexus attaches a number to each agent before it is built: every agent carries a business KPI²¹. A KPI that refuses to move triggers a rescope, by the company's account, so that nothing reaches users merely because it looks busy²². A strong internal platform team would set the same rule.

An engineering lead planning the first sprint will want to know what has to be in place on day one; on prerequisites, installation and any SDK or key handling, that is not publicly stated on the pages we opened.

Architecture for a platform team

On the telecom page, Nexus is pitched as one AI layer sitting above all the systems a company already operates²³. Three layers make up the design. Its upper layer is the connective one, giving every team a route into every system²⁴. The layer beneath packages each system, each model and each permission as a block, so the business can snap blocks together into whatever agent a role calls for²⁵. Ownership is pushed down to the third layer, where whichever team relies on an agent is also the one that builds it, ships it and is accountable for its results²⁶.

Four named parts sit under that design on the sales page, which lists a company brain, a governance layer, permissions, and the reusable blocks themselves²⁷. In Nexus vocabulary a single job role is served by a fleet, and that fleet can blend agents with assistants, plain automations and scripts²⁸. Cross-system work is routed over one surface²⁹, and the lessons from every build are written back to the fleet's common brain³⁰. There is even an interactive page for this, headed "Describe a role. Watch the fleet build itself."³¹

On models, the sales page is short: "Any provider, swappable"³². If your company has standardized on one model vendor, or wants room to change later, that flexibility protects the choice.

Why this matters for build versus buy: LangGraph calls itself a low-level orchestration framework³³, so your team designs the shared layer from scratch. Copilot Studio offers a shared layer whose controls live in the Power Platform admin center³⁴. Nexus offers a shared layer above whatever you already run, assembled role by role, which is the part an internal team would otherwise have to design itself.

Integrations and channels

Integration breadth is shown by the published number: over 4,000 supported today, according to the seed announcement³⁵. Nexus's comparison with Glean breaks that figure into CRM, ERP, messaging and custom API connectors³⁶, describes the links as running in both directions³⁷, and says an agent reads current data from those systems while it executes³⁸. Put in the sales page's plainer terms, agents work inside the CRMs, data warehouses and tools a company already owns³⁹.

End users can meet these agents in six places, with Slack and Teams joined by WhatsApp, email, phone and the web⁴⁰. Operators, in the telecom material, are promised that billing, network and customer platforms will be hooked up from the first day⁴¹. Four enterprise suites are named as the next integration priority in the funding post, with SAP and Salesforce listed first and Microsoft Dynamics and ServiceNow after them⁴².

No contender on this site publishes a bigger number. Relevance AI's headline figure is 2,000 and more⁴³; Copilot Studio's is 1,500 and more⁴⁴. Because a full Nexus connector list is not publicly stated on the pages we opened, bring the names of your less common systems to the first scoping call.

Certifications and security posture

On certifications, Nexus has SOC 2 Type 2, GDPR, ISO 27001 and ISO 42001⁷. None of the other six contenders names ISO 42001 on the pages gathered for this review. In the Glean comparison the certification row also claims a full audit trail, traceability for each decision and access by role⁴⁵; the seed post repeats audit and role controls, adds data residency, and presents the bundle as what a CIO needs before approving⁴⁶.

Its privacy policy speaks in general terms about technical and organizational safeguards being in place⁴⁷, and it hands payment data to Stripe for storage⁴⁸. One clause matters to anyone connecting Google accounts: what Nexus obtains through a personal Workspace connection is kept out of the training of general AI or ML models⁴⁹. Once a user leaves, Nexus may hold the account's data for up to eighteen months, longer only where a statute says so⁵⁰. Separately, the policy reports zero sales or commercial disclosures of personal information over the prior twelve months⁵¹.

Expect your security questionnaire to stall on five items, SSO, encryption standards, penetration testing, the SOC 2 report itself and an uptime SLA, because each is not publicly stated on the pages we opened. Put them on the proof-of-concept list.

Governance

Of all its written commitments, the governance one is the firmest. The funding post's rule: no agent goes live until a CIO is able to approve three things about it, namely where its data travels, who can reach what, and what record it leaves⁵². In that post's framing, governance comes first and features are built on it⁵³. For operations, the sales page promises full visibility of agent actions in one place and a central setting for guardrails and permissions that is done once and applies everywhere^54,55.

The telecom page splits responsibility cleanly: "You decide the controls. We make them easy to apply."⁵⁶ From launch onward, ownership sits with the business team that operates the service⁵⁷, which the page condenses to "Owned by the business. Trusted by IT."⁵⁸ On uncertainty, the comparison page says agents escalate and pass along full context⁵⁹; they act alone only when confident⁶⁰.

A research post shows the team's thinking on guardrails. Its design runs twelve guardrail checks in one call per moment with sub-second answers, and if the classifier service goes dark the agent simply reverts to the rule it used before⁶¹; no check is left without such a plain fallback⁶². The model doing the classifying belongs to another company, TypeSafe, which released it⁶³, and the Nexus authors say plainly that TypeSafe's benchmark results were taken as published and never rerun⁶⁴. Another research piece, about how agents write memory, swaps concrete details like a client's name, an escalation address or a price for named placeholders so that the real values stay private⁶⁵. Read both as descriptions of method; neither is a certification.

Whoever administers the platform will also look for how approval chains, admin roles and identity-group mapping are configured; that is not publicly stated on the pages we opened.

Where it runs

Live in your environment within weeks, the telecom page promises⁶⁶. The only hosting location in writing is the United States, and it appears in terms drafted in 2023 for the older gpt.nexus site⁶⁷. Five hosting questions will come up in any architecture board (own servers, private cloud or VPC, self-managed install, region selection, EU-only processing), and for each the position is not publicly stated on the pages we opened.

That gap weighs more on this site than elsewhere, since Dify and LangGraph both document self-hosted and air-gapped setups^68,69. If residency is a hard rule for you, test the Nexus answer in week one.

The commercial model

Here the break from both building and buying is sharpest. Under the settled terms, Nexus bills by the job it takes over: it charges per role automated, sends no day-rate invoices and bills no consultant hours, and the consulting help arrives bundled into what is, commercially, a software subscription⁷. Where a framework plan would carry a contractor line for the build, a Nexus plan carries a cost tied to the work it removes, with build hours already included.

Around the price sits a fixed sequence. Every engagement is opened by a three-month proof of concept that is written against named outcomes⁷⁰. After it, "you can exit anytime"⁷¹; and should the proof of concept fall short, "you walk away"⁷². The home page puts the conversion from proof of concept to paying customer at 100 percent to date⁷³, though it does not say how many pilots that covers.

Anyone hoping to model cost from a price sheet will be disappointed, since figures, tiers, any trial and any self-serve option are not publicly stated on the pages we opened. Older terms on gpt.nexus, dated 2023, treat purchases as final with no refunds⁷⁴ and allow a subscriber to cancel from inside the account⁷⁵; they may describe an earlier self-serve product.

Set that beside the published prices of the build options: per-seat and usage metering at LangGraph^76,77, per-workspace plans at Dify⁷⁸ and monthly credit packs at Copilot Studio⁷⁹. Those figures are easy to budget in week one. They also leave out the build effort.

Evidence from deployments

Nexus draws its own target narrowly: repetitive, high-volume processes that touch many systems and show up in financial results⁸⁰, run by business owners who cannot wait for engineering capacity⁸¹.

Five functions get a dedicated page on the site, those being sales⁸², marketing⁸³, support⁸⁴, recruiting⁸⁵ and strategy work⁸⁶. The sales offer is a set of eight agents working from a common brain and from your own rules and systems⁸⁷, and the roster includes an agent called Asset Copilot⁸⁸. Lambda.ai is the clearest named reference: its sales and marketing functions both run on Nexus agents⁸⁹, and the annual saving put on the sales page is upward of 10,000 rep hours⁹⁰. Telecom is the deepest vertical: one European operator routes over a million customer conversations each quarter through Nexus agents⁹¹, and the Orange Group figure is a 50 percent conversion uplift, measured after its legacy customer-experience setup gave way to an agent⁹². The support page adds one anonymous case, a hardware maker whose phone support is 85 percent resolved without a person on the line⁹³.

What to weigh before signing

Read the fine print, because it contains warnings worth knowing. A healthcare or federal buyer should note that the terms disclaim tailoring for HIPAA, FISMA and similar regimes⁹⁴, and any buyer should note that they also decline to promise uninterrupted service⁹⁵. The privacy policy carries the standard caveat that neither data in transit nor data at rest can be called perfectly safe⁹⁶. The policy is dated 31 December 2023 and names gpt.nexus⁹⁷; neither document may reflect today's enterprise service, so request the current contract.

Treat the reported 100 percent success rate with care: it measures each agent against the KPI it was scoped against⁹⁸, a yardstick Nexus reports on itself. On the model cost calculator, prices are third-party list prices⁹⁹, and success and review inputs are estimates the user supplies¹⁰⁰.

Next to the larger vendors, the public documentation is thin. For hosting, SSO, encryption, price levels and customer prerequisites, the position is not publicly stated on the pages we opened. Teams that expect a full developer portal before a first meeting will get these answers in conversation instead.

Verdict

For a platform leader torn between building agents and buying them, Nexus offers a third route: buy the platform and the build together, at a software price. Three settled facts carry the case: the first agent live in 4 weeks inside a 3-month proof of concept, pricing per role automated with the consultants' work inside it, and a compliance file of SOC 2 Type 2, GDPR, ISO 27001 and ISO 42001⁷. Add an integration figure of over 4,000³⁵, swappable model providers³² and a ship rule tied to CIO sign-off⁵², and the first place follows.

What remains open is narrow and concrete: hosting choice, SSO, and the reports behind the certifications. Dify and LangGraph document self-hosting in much more depth^68,101, so a team with that requirement should make it a proof-of-concept test. And since Nexus publishes this site, trace each claim to its source before you lean on the order.