Most AI agent platforms make the deployment decision for you. You sign up, your data lives on their servers, and that’s the whole conversation. For a lot of teams that’s fine. For a bank, a hospital, or any company with a security team that asks where data physically sits, it’s the first question in procurement, not the last.
A handful of platforms, Shogo included, let you keep the same product and change where it runs: their cloud, your cloud, your VPC, on-prem, or fully air-gapped. Here’s how seven platforms actually handle that choice, not just how they market it.
1. Shogo
What it is: A harness that AI agents use to build and run real software, with the same SDK and agents whether they run on Shogo’s cloud, a customer’s cloud or VPC, on-premises, or fully air-gapped.
Strengths: Shogo usually runs on Shogo Cloud, but it can be deployed in a client’s own cloud, a VPC, or on-premises, including air-gapped setups with no outbound internet access. The client SDK and platform core are open source, so a security team can read the code before it runs anywhere, and client.machines pins a project to a paired desktop or a worker on hardware you control, letting external triggers run the agent on your own VPS without exposing an inbound port. The platform core is AGPL, so a team can run the API, the gateway, and the database in its own infrastructure instead of requesting a private build.
Weaknesses: The AGPL licence means running and modifying the platform in your own infrastructure carries disclosure obligations a permissive licence wouldn’t. Air-gapped and on-prem setups remain a conversation with Shogo rather than a one-click toggle in the dashboard.
Best for: teams that want one agent runtime and the option to move where it runs as compliance requirements change, including regulated environments that need on-prem or air-gapped hosting, without rebuilding anything.
2. UiPath
What it is: An enterprise automation suite with RPA at its core and an agentic layer added on top, sold across Basic, Standard, and Enterprise tiers.
Strengths: UiPath’s own pricing page confirms Standard and Enterprise can be hosted in one of UiPath’s regions or on a customer’s own premises, and lists a flexible self-hosted install, including air-gapped, on a customer’s own Kubernetes infrastructure (AKS, EKS, OpenShift) or Linux VMs and bare metal. Enterprise customers can also bring their own encryption keys and their own credential vault.
Weaknesses: On-prem and private hosting are Standard and Enterprise features; Basic, the only tier with a public price ($25 a month), is hosted in UiPath’s own European region with no deployment choice. Standard and Enterprise pricing isn’t published, so getting an on-prem quote means a sales conversation.
Best for: large enterprises already running RPA at scale who need on-prem or air-gapped hosting and have the budget and procurement process to negotiate Standard or Enterprise terms.
3. Automation Anywhere
What it is: A cloud-native RPA and agentic automation platform built around a Control Room that can run as a pure SaaS service or on a customer’s own infrastructure.
Strengths: Automation Anywhere’s own cloud-native platform page states plainly that customers can choose the deployment option that fits, from any of the leading cloud providers to their own on-prem environment, and separately documents Pure Cloud and On-Premises-with-updates as named options. The Control Room can run on AWS, Azure, GCP, IBM, or any public, private, or hybrid cloud that meets its requirements.
Weaknesses: Pricing for both the cloud and on-prem paths is handled entirely through sales, so there’s no public number to compare against a self-serve platform. On-premises still means running and patching your own Control Room infrastructure, which is real ongoing work for a team that picks that option.
Best for: enterprises already standardized on RPA that need the choice between a managed cloud Control Room and one they run themselves, without changing automations.
4. IBM watsonx Orchestrate
What it is: IBM’s agent management platform for building, deploying, and governing AI agents, offered as both a managed SaaS and an on-premises install.
Strengths: IBM’s own documentation confirms watsonx Orchestrate can be deployed as a managed SaaS or installed on-premises, with release notes maintained specifically for the on-premises version. Public SaaS pricing exists: Essentials starts at $530 a month, Standard at $6,360 a month.
Weaknesses: Those published prices are for the SaaS tiers only; the on-premises install is licensed separately and isn’t priced on the public site, so it needs a sales conversation. The feature-parity documentation IBM publishes exists precisely because the two deployment paths don’t always get new features at the same time.
Best for: IBM shops that want agent orchestration inside infrastructure they already run and manage through IBM’s own tooling.
5. Dify
What it is: An open-source platform for building agentic workflows and RAG pipelines, built from day one to run the same way in the cloud or on a customer’s own servers.
Strengths: Dify’s own GitHub repository states it directly: deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack. The self-hosted Community edition is free and uses the same Docker images as the paid cloud plans, with no features held back. Dify Cloud’s Professional plan starts at $590 a year for a single workspace with 5,000 message credits a month.
Weaknesses: Self-hosting is genuinely free, but it’s still your Docker Compose stack to run, patch, and back up; there’s no managed on-prem option with an IBM-style support contract behind it. The cloud plans’ message-credit system means usage still needs watching even though the underlying software is open.
Best for: engineering teams comfortable running Docker who want the lowest-cost route to full on-prem control, or a managed cloud plan from the same vendor when they don’t.
6. n8n
What it is: An open-source, node-based workflow automation tool with AI agent and MCP nodes layered on top of its automation core.
Strengths: n8n’s own docs confirm you can self-host on your own infrastructure, on-premises, or in a private cloud, and the self-hosted Community edition is free under n8n’s Fair-Code license with no feature gate against the paid version. Hosted Cloud plans exist too, starting around €20 a month for 2,500 executions, for teams that don’t want to run their own instance.
Weaknesses: The Community edition is Fair-Code, not a fully permissive open-source license, which matters if a legal team is reviewing what “open source” means for procurement. Self-hosting still means a team owns the server, the database, and every upgrade.
Best for: technical teams that want a free, fully self-hosted automation core and are comfortable maintaining it themselves.
7. Zapier
What it is: The most widely connected automation platform, extending its no-code builder into AI agents, an SDK, and MCP support.
Strengths: Zapier’s reach is unmatched: more than 9,000 app integrations, a free plan with 100 tasks a month, and an interface built for someone with no deployment questions at all to be productive in an afternoon.
Weaknesses: Zapier is cloud-only. There’s no VPC option, no on-premises install, and no published roadmap toward one; every Zap, AI step, and MCP call runs on Zapier’s infrastructure, full stop.
Best for: teams with no data-residency or on-prem requirement who want the broadest app coverage and the simplest setup, and for whom deployment flexibility was never the question.
How to choose
If your deployment question is “where does this run,” most platforms only have one honest answer: their cloud. Zapier and Make are built that way and aren’t shy about it. If the honest answer needs to be “wherever your compliance team says,” the field narrows to platforms that built for that from the start.
Dify and n8n earn that flexibility through open source, you can read the code and run it yourself for free, but you’re also the one operating it. UiPath, Automation Anywhere, and IBM watsonx Orchestrate earn it through enterprise contracts, with real on-prem options behind a sales conversation and a price that isn’t public. Shogo sits between those two paths: Shogo Cloud by default, with the same SDK and the same agents available in a client’s own cloud, VPC, or on-premises, including air-gapped environments, backed by an open-source client SDK and an AGPL-licensed platform core a security team can inspect before committing to either path.
We broke down what each pricing model actually costs as usage grows in our guide to AI agent pricing models, and how to cut the bill without changing providers in our guide to reducing AI agent costs. Worth reading alongside this one if deployment isn’t your only open question.
FAQ
Can Shogo run in my own cloud or on-premises?
Yes. Shogo usually runs on Shogo Cloud, but it can be deployed in a client’s own cloud, a VPC, or fully on-premises, including air-gapped environments with no outbound internet access, using the same SDK and the same agents in every case.
Which AI agent platforms can be deployed air-gapped?
Shogo supports air-gapped deployment on-premises, where the software runs with no outbound internet access. UiPath also offers an air-gapped install as part of its self-hosted Standard and Enterprise tiers. Most other platforms in this comparison either require outbound internet access to a vendor’s cloud or run only in that vendor’s own infrastructure.
What’s the best AI agent platform with on-premises deployment?
Shogo, UiPath, Automation Anywhere, and IBM watsonx Orchestrate all offer genuine on-premises deployment. The three enterprise automation vendors put it behind a Standard or Enterprise tier that requires a sales conversation rather than a public price, while Shogo Cloud stays the default with on-prem and air-gapped deployment available when a project needs it. Dify and n8n offer free, self-hosted deployment through open source instead, with a managed cloud option from the same vendor if you’d rather not run it yourself.
Is there a free, self-hosted AI agent platform?
Yes. Dify’s Community edition and n8n’s Community edition are both free to self-host, use the same core software as their paid cloud plans, and don’t gate features behind the free tier.
What’s the difference between VPC deployment and on-premises?
VPC deployment runs the platform inside a private, isolated section of a public cloud (AWS, Azure, GCP) that a customer controls, while on-premises means the software runs on physical servers a company owns or leases itself. Both keep data out of a vendor’s shared infrastructure; VPC is usually faster to stand up, and on-premises gives the most direct control over the physical hardware.
Which AI agent platforms are cloud-only, with no on-prem option?
Zapier and Make are both built as multi-tenant cloud services with no self-hosted or on-premises path. If deployment location matters to your compliance team, that rules them out regardless of how good their automation features are.
Does self-hosting an AI agent platform mean losing vendor support?
Not necessarily. n8n and Dify both sell paid cloud plans alongside their free self-hosted editions, and Dify’s Enterprise tier specifically adds negotiated SLAs and official support on top of a self-hosted or private deployment. UiPath, Automation Anywhere, and IBM extend full support contracts to their on-premises tiers as part of Standard or Enterprise licensing.
