The NewCloud Tech Stack

Sovereign serverless cloud for AIware apps, created by an open, mathematical network — empowering governments, enterprises, and people to build a secure online future with AI.

Ecosystem spotlight · All markets

Build anything on the Internet Computer

Explore apps, emerging platforms, and AIware, built on the Internet Computer.

Scroll position 1 of 1, 7 apps in total.
Agentic infrastructure

No infrastructure security team. No sysadmin team.

Tamperproof

Apps are tamperproof network code. No app servers to break into, database servers to compromise, or firewalls to maintain.

Always on

Apps run as serverless code, without database servers or traditional infrastructure to go wrong. The network keeps them running and their data available.

Safer upgrades

Agents can safely improve live apps. Language frameworks like Motoko check that updates preserve existing data, and reject incompatible changes that could cause accidental data loss.

Sovereign control

Cloud engines let you precisely choose where your compute capacity runs and who operates it, while your apps keep running — delivering dynamically configurable sovereignty and complete freedom from vendor lock-in.

Gain the freedom to put your agents to work without speed bumps.

See how the network runs apps

Safe autonomy for agents.Much more intelligence in apps.

The Internet Computer turns an open network into a cloud where agents can build, host, and evolve apps — and work inside them.

2.5 Software AI can work inside

Meet AIware.

AI chat becomes a new way to use your software, working alongside its traditional screens. Analyze data across apps and orchestrate complex tasks through conversation, within your permissions.

Open SaaS AssistantExample request
“Review all Customer Success time-off requests. Approve those that preserve coverage in every time zone, and suggest alternative dates where requests conflict.”

One conversation can draw on Open HR, Open CRM, Open Email, and the enterprise’s wider data graph. The Assistant analyzes data and coordinates actions across the suite, using only the data and functions your permissions allow.

Select an agent to see what it can access.

Your AIware apps and servicesBuilt with Motoko 2.0
OQL APIAgents analyze in-memory data graphs
App functionsAgents perform complex tasks hands-free
Permissions are part of the program.Structures, records, and fields. Scoped to the agent or the user behind it.

What this caller can access

HRQueryable via OQL
CRMQueryable via OQL
AccountsQueryable via OQL

CEO chat can query HR, CRM, and Accounts. All memory groups in this example are accessible.

Each app’s permission rules can only change by rebuilding that app.

Through Object Query Language (OQL), agents query the data each app permits them to see, and use callable functions to act. Each app enforces its own permissions.

Languages for AIware

Motoko

The first programming language built for agents.

A language designed for agents to safely and rapidly build and evolve enterprise apps, services, and infrastructure, and imbue them with AIware capabilities.

WebAssembly

An open standard for compute.

The Internet Computer runs WebAssembly (Wasm), a low-level bytecode format for fast, portable execution. Supported by all major browsers and used across backend and cloud platforms, it is an open W3C standard.

Any language that compiles to WebAssembly can target the Internet Computer. Motoko and Rust are the two most popular.

2.6 Software agents can build

The program is the database.

Data lives with your code in persistent memory. Agents can focus on the essential logic of your application, with less code to generate and maintain — and more scope to build sophisticated software.

  • Data within reach.Code works directly with data in persistent memory, avoiding round trips to a separate database.
  • Less code. Less generation cost.Removing storage plumbing means fewer tokens for agents to generate, helping reduce development time and cost.
  • More ambitious applications.A simpler program model helps agents reason about complex relationships and behavior.

Application logic+Persistent data

One program. Hosted by the network.

A tiny inventory appMotoko
persistent actor {
  // Persistent data
  let inventory = {
    var stock : Nat = 10;
    var orders : Nat = 0;
  };

  // Application logic
  public func order() : async () {
    inventory.orders += 1;
    inventory.stock -= 1;
  };
}

Declare it. It persists.

inventory and the whole data structure it references are stored automatically, between calls and across upgrades. No loading. No saving.

Both changes. Or neither.

Each order() updates memory as one transaction. With no stock left, execution traps and every change is rolled back — even the order count. No half-written data.

If two orders arrive together, each runs against the latest inventory, so one cannot accidentally overwrite the other’s changes.

Why this helps AI build

Motoko’s orthogonal persistence preserves persistent variables and the data structures they reference across calls and upgrades. Code works directly with this data, without translating it into database queries or file formats.

Fewer output tokens. Database queries, data conversion, and storage plumbing all add code. Removing that work means fewer code tokens for an agent to generate, which can reduce generation time and cost.

A simpler model to reason about. Orders, customers, and their relationships can be expressed directly as connected data structures. Agents can focus more of their reasoning on the app’s behavior, helping them build and maintain more sophisticated software.

In this example, order() has no await, so each call runs as one transaction. Work spanning asynchronous calls needs its own coordination. Read about execution boundaries.

How ICP runs apps.

Your app’s code and data run together in a canister, copied across its subnet. A malicious node alters its local purchase-order amount, but the healthy nodes execute the authorized update. Your app verifies the network certificate for the correct result. The forged record fails verification. A separate example shows the app running through three node outages. Your app can also call services on other ICP subnets; ICP routes their requests and replies.

Compute capacity, from independent providers

Specialized hardware, and hyperscaler compute capacity, all over the world.

This 13-node example tolerates up to four faulty nodes. Progress needs timely communication. Record integrity requires certificate verification; an unverified query is not enough. How verification works

The self-writing cloud

The Internet Computer provides next-generation agentic infrastructure

4.1

Software is reimagined for agents

AI agents build vastly faster than humans can, and they have their own needs. The Internet Computer runs software inside persistent memory, introducing a powerful new paradigm where the program is the database. Now a program's essential logic persists data automatically, reducing complexity, and fueling the modeling power of AI, enabling AI to go further, faster, and at less cost.

4.2

Mistakes made at speed can't lose data

AI agents can develop apps and enterprise infra at unprecedented speed. However, upgrading traditional production systems is risky, as mistakes can result in data loss. On the Internet Computer, new languages like Motoko automatically verify the safety of every upgrade, preventing mistakes resulting in data loss, so you can remove the brakes.

4.3

No infrastructure oversight necessary

Traditional infrastructure is insecure and sensitive to components such as databases being misconfigured or failing. Apps on the Internet Computer are tamperproof, keep their data in persistent memory, and are guaranteed to run. This means apps don't need support from security and sysadmin teams, and you can give your agents far more autonomy.

The research program

The world's most sophisticated network, is yours.

More than half a billion dollars has been spent developing the technology that powers the Internet Computer, which represents the most sophisticated network the world has ever seen. The purpose of the project is to extend the internet, so that it can host software, as well as connecting software.

In recent years, the project has focused on developing the Internet Computer as agentic infrastructure. The Internet Computer unlocks the full potential of AI by force-multiplying agents that build apps, enabling the creation of AIware software that integrate AI much more closely than traditional software can, and removing security and sysadmin requirements, so agents can be allowed to work with far greater autonomy.

The latest innovation, "cloud engines," enables the creation of private subnets that can be arbitrarily configured, for example to combine compute hardware located in the home country, or operated by specific providers, unlocking the simple creation of fully sovereign tech stacks that meet national security needs, or regulatory needs such as GDPR compliance. Cloud engines keep on running even when the underlying compute capacity is swapped, making vendor lock-in history.

Research and resources

The research behind the Internet Computer

Explore the papers and presentations that explain how the Internet Computer works, why it was designed this way, and what it makes possible.

In the press

Recent coverage, announcements, and field notes.

Ecosystem spotlight · Government & NGO

Institutions building on the sovereign-by-default stack.

Quick start · for human developers

Everything you need to ship apps, websites, and services inside the Internet Computer's tamperproof serverless environment.

Start building
For agents
.md ↓.json ↓