DwiziDwizi

Browse docs

Concepts

Tools, runners, MCP servers, tokens, and visibility.

Core Concepts

Dwizi is built around a small set of primitives that make AI tooling safe, predictable, and easy to operate at scale.

These concepts apply whether you are creating your first AI tool or running production systems across an organization.

Tools: Executable Capabilities for AI

A tool is a unit of execution that gives an AI system the ability to act.

Examples:

  • Fetch live data
  • Query internal systems
  • Transform files
  • Trigger external APIs

A tool is not a script or a service. It is explicit logic with defined inputs and outputs.

Every Dwizi tool consists of:

  • Logic — a JavaScript or TypeScript function
  • Interface — a clear input/output schema
  • Isolation — executed in a secure, independent runtime

Tools are designed to be:

  • Deterministic
  • Stateless by default
  • Safe to call by AI systems

Runtimes: Isolated Execution Environments

A runtime is where a tool executes.

Dwizi provides hardened runtimes so you never manage servers, containers, or scaling.

Currently supported runtimes:

  • Deno (recommended)

    • Native TypeScript
    • URL-based imports
    • Security-first execution

  • Node.js

    • Compatibility with existing JavaScript code
    • Familiar APIs

Runtimes are:

  • Created on demand
  • Isolated per tool
  • Automatically cleaned up after use

Interfaces: How AI Systems Use Tools

AI systems interact with tools through standardized interfaces.

MCP (Model Context Protocol)

Dwizi tools can be exposed via Model Context Protocol (MCP), allowing AI systems to:

  • Discover available tools
  • Understand tool inputs and outputs
  • Execute tools reliably

An MCP server represents a collection of related tools exposed through a single interface.

This means:

One connection gives an AI system access to multiple capabilities.

MCP is one supported interface. Dwizi is designed to support additional tool interfaces as AI ecosystems evolve.

Access & Security

Tokens

Access to tools is controlled via tokens.

  • Private tools require authentication
  • Public tools can be shared openly

Tokens define:

  • Who can access a tool
  • Which environments it can be used in

Organizations

All resources in Dwizi belong to an organization.

Organizations provide:

  • Team collaboration
  • Centralized access control
  • Usage visibility
  • Resource boundaries

This allows AI tooling to be governed like any other enterprise system.

Retention & Audit

Audit events are retained by default. Output/log retention is configurable and can be disabled or obfuscated for compliance-sensitive workloads. See Data Retention.

Execution Model (End-to-End)

Logic → Tool → Runtime → Interface → AI System
  ↓       ↓       ↓          ↓
Function Schema Isolation  MCP / API

Dwizi sits at the center, ensuring that execution is:

  • Safe
  • Observable
  • Repeatable

Why These Concepts Matter

For Builders

  • Write logic, not infrastructure
  • No servers, no deployment pipelines
  • Clear contracts for every tool

For AI Systems

  • Reliable execution
  • Explicit capabilities
  • Predictable behavior
  • Safe isolation

For Organizations

  • Governed AI tooling
  • Auditable usage
  • Strong security boundaries
  • Cost and access control

Getting Started

  1. Define the tool's logic
  2. Declare its inputs and outputs
  3. Choose how it is exposed
  4. Connect it to an AI system

Dwizi handles execution, isolation, and lifecycle management automatically.

Next Steps