Nuphos Review (2026): Pricing, Features & Honest Verdict

Nuphos is a DevOps workspace where AI agents investigate and operate your cloud infrastructure under IAM-scoped permissions and approval flows. The free desktop console is good. The paid tiers meter an allowance the page never defines, and one is priced worse than the tier below. Best for small infra teams. Price: from $29/mo, free tier yes. Rating: 7/10.

What is Nuphos?

Nuphos is a shared environment where engineers and AI agents work on the same infrastructure from the same place. Instead of an agent running loose in a terminal with your production credentials, Nuphos scopes each agent session to a native IAM role, routes write actions through approval flows, and writes everything to an audit trail. The default is read-only. The company describes it as bringing “the power of terminal agents, tamed for production DevOps,” and TAME is the actual framing: Trust, Action, Memory and Experience.

We found it on Product Hunt where it launched as “The AI-Native DevOps Workspace.” Underneath the agent story there is a straight cloud console covering AWS, GCP, Azure, Kubernetes, Cloudflare and Zeabur, with the full Kubernetes object tree from pods and deployments through to CRDs, network policies and RBAC bindings. The team is small, under ten engineers, with Yuanlin Lin and Ling Wu among the makers. Note that fmerian appears on the maker list and is a prolific Product Hunt hunter rather than part of the company.

What are Nuphos’s key features?

IAM-scoped agent sessions

This is the feature that makes the rest defensible. Every agent session is bound to a real IAM role from your cloud provider, not a Nuphos abstraction sitting on top of one. The site shows sessions scoped to AWS ReadOnlyAccess, a narrower deploy-staging role, GCP logging.viewer and Azure KeyVault Admin. Because the permission boundary is enforced by the cloud rather than the app, an agent that misbehaves is constrained by infrastructure you already trust and already audit.

Approval flows and audit trails

Write actions can be held for human approval, with the request showing the target, the environment, the role and the change window before anyone clicks yes. Every action lands in an audit log. The pattern is familiar from change management tooling, and applying it to agents is the correct instinct: the interesting failure is not an agent reading the wrong log, it is an agent restarting the wrong service at the wrong time.

Infrastructure memory

Nuphos builds a persistent picture of your specific environment, storing services, environments, workflows and operational history so agents are not rediscovering your architecture on every invocation. The demo shows learned facts accumulating across api-server, postgres, prod, deploy flows, incident records and runbooks. This is the part that should compound, and it is also the part that is hardest to evaluate in a trial, because a memory layer is worth little on day one and a lot in month six.

Agents where your team already talks

The agent is reachable from Slack, Discord, Lark and iMessage, so kicking off a migration, approving an action or checking an incident does not require opening another console. For on-call work this is the difference between a tool people use at 3am and one they do not. There is also a Vanta and Secureframe integration that turns SOC 2 and ISO 27001 findings into remediations rather than a list of things to fix later, which is the most concrete use case on the site.

How much does Nuphos cost?

Free is $0 and gives the full console with one cloud connector and up to three members, but no agents. Dev is $29 a month, Team is $99, Business is $299 and Scale is $999. Everything is priced per workspace rather than per seat, and from Team upward members are unlimited, which is a real differentiator against per-seat DevOps tooling. Team gets up to three cloud connectors, Business up to ten, Scale unlimited.

Direct comparison is difficult because the obvious competitors do not publish prices. Resolve.ai routes to a contact form. incident.io publishes a pricing guide rather than a rate. Against those, publishing four committed figures is a point in Nuphos’s favour, and $29 a month for an agent that can investigate an incident is inexpensive against an engineer’s hourly cost.

What does “agent usage” actually mean?

Here is the problem. Every paid tier is differentiated primarily by agent usage, and agent usage is expressed only as a multiple of the Dev plan. Team gives 5x the Dev plan’s agent usage. Business gives 10x. Scale gives 50x. At no point does the page say what 1x is.

Not in tokens, not in tasks, not in dollars of model spend, not in hours. The closest the page comes is the line “priced like Claude Code and Codex,” which describes a philosophy rather than an allowance. You cannot size a plan against your own workload, you cannot forecast a bill, and you cannot tell whether the Dev plan covers one incident a month or fifty.

To be fair to Nuphos, they handle overage well. Team and above are pay-as-you-go beyond the included usage, with a monthly cap you control on Team, and the site commits that “limits never interrupt an incident” and that monitoring does not count against agent usage. Those are the right policies. They just do not substitute for a number, because a cap you control is only useful if you know what you are capping.

Which plan is the wrong one to buy?

Business, on volume grounds. Divide each price by its stated multiple of Dev usage and the ladder does something odd.

Dev is $29 for 1x, so $29 per unit. Team is $99 for 5x, which is $19.80 per unit. Business is $299 for 10x, which is $29.90 per unit. Scale is $999 for 50x, back down to $19.98 per unit. Business is the worst per-unit deal on the entire page, worse than Team by more than 50%, and marginally worse than the $29 entry plan.

Put another way: three separate Team workspaces would cost $297 and carry 15x the Dev plan’s agent usage, while one Business workspace costs $299 and carries 10x. You would pay two dollars less for half as much again.

There are real reasons to buy Business anyway. It lifts cloud connectors from three to ten, adds extended audit history, and cuts support response from one working day to four hours. If you need ten connectors, you need ten connectors. But the page presents the ladder as though usage scales with price, and on usage alone the step from Team to Business is the one place where paying more buys you less per dollar. Check the connector count you actually need before assuming Business is the natural upgrade.

Does Nuphos only run on macOS?

The pricing page says so twice. The headline reads “Free forever as the all-in-one macOS workspace for your DevOps team,” and the Free plan card repeats it. Read that and you would conclude this is a Mac-only product, which for a team tool would be a serious limitation given how many infrastructure engineers work on Linux.

It is not true. The download page offers six builds: Apple Silicon, Intel Mac, Windows x64, Windows arm64, Linux x64 and Linux arm64. The product is cross-platform and the pricing copy is understating it, which is an unusual direction for marketing to be wrong in.

One caveat from that same page, and credit to Nuphos for stating it: the Windows arm64 build is marked unsigned. Expect SmartScreen warnings and a slightly awkward install for anyone on a Surface or an arm64 VM. Everything else appears to be signed and ships as a normal installer, with macOS delivered as a .dmg.

Who is Nuphos best for?

Use Nuphos if you are a small infrastructure team, roughly three to ten engineers, already running on AWS, GCP or Kubernetes, and you want agents involved in operations without handing them unscoped credentials. The unlimited-members pricing from $99 makes it cheap for a whole team rather than a per-seat decision, and the compliance remediation path through Vanta or Secureframe is a concrete win if you are mid-certification. The free tier is a legitimate console in its own right, so evaluating costs nothing.

Skip Nuphos if you cannot buy against an undefined unit, which is a reasonable position for anyone with a procurement process. Skip it if you are a large enterprise, because ten connectors and a four-hour support SLA is not an enterprise posture and there is no custom tier above Scale. And skip it if your agents need to run unattended at scale, since the value here is in supervision and approval rather than autonomy.

Best Nuphos alternatives

Resolve.ai targets the same AI-for-production-operations space with a stronger enterprise story. It publishes no pricing at all and routes everything through a sales conversation, so it is a different kind of purchase.

incident.io is the mature option for incident response with a large installed base and deep Slack integration. It is stronger on process and weaker on agents actually operating infrastructure, and it also publishes a pricing guide rather than public rates.

Claude Code or Codex with your own kubectl access is the zero-cost baseline and what most teams do today. You get no IAM session scoping, no approval queue, no audit trail and no infrastructure memory, which is precisely the gap Nuphos is selling into.

Final verdict: is Nuphos worth it?

The engineering judgement here is good. Scoping agent sessions to native cloud IAM roles rather than inventing a permission model is the right call, approval flows and audit trails are the correct controls, and pricing per workspace with unlimited members is a rare and friendly structure in a market that loves per-seat billing. The free tier is a usable console, the desktop app ships for three operating systems, and the compliance remediation use case is sharper than most of what launches in this category.

What stops it scoring higher is that you cannot work out what you are buying. The one thing separating the paid tiers is measured in multiples of an allowance that is never stated, and when you do the arithmetic on those multiples, the $299 plan turns out to be the worst value on the page. Both problems are fixable with a paragraph of copy. Until then, start on Free, move to Dev at $29 to see how far the usage goes on your own workload, and jump to Team rather than Business unless you specifically need the connectors.