CloudTop Platform

The agent platform, in technical terms.

AI agents need three things to do real work: a real runtime, live app context, and governance that ships with them. Here is how CloudTop delivers each one — today.

The thesis

Software built for both
humans and AI.

One artifact, two faces: humans see the interface; AI sees structured tools and live state through an on-device MCP server — under the same permissions as its user.

How it works

AI that works inside your apps —
not from the outside looking in.

Most AI reaches your tools from outside — an API or server, data in transit. CloudTop runs the connection on the client, inside your running apps. Nothing leaves.

The usual way

AI from the outside

  • × Reaches apps through a server or API endpoint
  • × Sees only what the API chooses to expose
  • × Your data and tokens travel to a middle service
  • × Blind to what the app is actually doing right now
CloudTop

AI from inside, on the client

  • Connects inside the running app — client-side MCP
  • Reads the app's live runtime state, not a summary
  • Acts with your credentials — they never leave the device
  • Orchestrates across apps running side by side
Request lifecycle

One prompt.
Every app. Zero integration.

How a request flows through CloudTop
User intent becomes coordinated action.
User intent
Organise my calendar for next week, summarise unread emails, and update the CRM.

CloudTop

Agent orchestration
Key ring & permissions
Model routing
Runs client-side
User's connected services
Calendar
Email
CRM
Files
The checklist

Six things an agent platform
must deliver.

Six questions any organization should ask before letting AI act — and what CloudTop ships for each one.

1 What's needed

A real permission & runtime model

Allow-or-deny is too crude for agents that chain actions. Permissions must be expressible, revocable, and auditable — for people and AI.

What CloudTop ships

Schema-verified apps, per-object ACLs

Every app is a declarative document, validated before execution. ACLs scope each person and each AI to app data, shared knowledge, prompts, and documents.

2 What's needed

Containment

Agents that act need isolation — somewhere mistakes and manipulation can't reach the rest of your world.

What CloudTop ships

Sandboxed, on-device execution

Agents run in a sandboxed runtime on your device and act through each app's typed MCP tools — never raw OS control, no rented cloud desktop.

3 What's needed

Security that ships with the capability

Prompt injection now turns into real actions. Controls have to arrive together with the agents — not years after the first wave of abuse.

What CloudTop ships

Client-side credentials, bounded blast radius

The AI authenticates as its user; tokens never leave the device. Schema validation and row-level ACLs cap what any injected prompt can do.

4 What's needed

Reach beyond one operating system

If AI lives in browsers, phones, and SaaS, an agent platform tied to a single OS is a bet against where work is going.

What CloudTop ships

One codebase, native on every OS

The same app runs native on Windows, macOS, Linux, mobile, and web via the marketplace — with data under European jurisdiction.

5 What's needed

Tooling for builders and agents

Builders need to create, supervise, and trust agent-era software — without being locked into one vendor's stack.

What CloudTop ships

Any AI tool, your existing database

Generate the front end with any AI tool and bind it to the database you already run. Apps arrive AI-native; agents and tools extend them.

6 What's needed

Collaboration without model lock-in

Vendor team plans solved sharing — and created lock-in. The team's work doesn't follow the model: switch LLMs, and the context stays behind.

What CloudTop ships

Portable AI memory, per-user model choice

Apps are multi-user by default. Teams share files and AI memory as portable artifacts while each member picks their model. One memory. Any model.

From any AI tool to every screen

Build your app with any AI tool. CloudTop makes it AI-native, sovereign, and native on every device.

1

Build with any AI tool

Create the front end with whatever AI builder you like — no lock-in to one tool.

2

Connect your existing data

Point it at the database you already run — no backend rebuild.

3

AI-native & sovereign, automatically

Instantly AI-native — connected to your models and apps. Credentials stay on the device.

Your AI modelsYour dataYour apps
4

Distribute everywhere

Publish to the marketplace — native on Windows, Mac, Linux, and mobile.

Get early access

Already built. Soon ready for you.

CloudTop is soon rolling out. Leave your email and we'll be in touch as access opens up.