The Technology Behind ZPT

How AI agents work, what ZPT builds on, and how it connects to your existing tools.

How an Agent Works

There's been a lot of talk about AI agents. But few companies know how to actually implement them. ZPT bridges that gap.

Agent

The Brain

The AI model understands your company context and decides what to do next.

The Hands

MCPs, plugins, and sub-agents that execute: read files, search the web, update your CRM.

Claude Code / Codex CLI / Gemini CLI = the “harness” (what is an agent?)

Your Company Setup

The agent works from your company’s information, workflows, and connected tools.

ZPT provides a battle-tested setup and personalizes it for your company.

Your team accesses it through Claude Desktop, ChatGPT, or any compatible app.

Built on Claude Code and Codex

ZPT setups are built on the same architecture that developers use worldwide: a structured local folder that gives AI full context about the company. Claude Code and OpenAI Codex are the engines that power the agent. It reads the folder, executes tasks, and connects to external tools.

+The folder stays on the company's machine. Nothing is sent to ZPT.
+The AI provider is interchangeable: Claude, ChatGPT, or open-source models
+ZPT configures everything. The company just uses it.

Claude Desktop

Codex Desktop

We help your team get set up and running. No technical knowledge required.

What ZPT Builds

The skills ZPT brings to every engagement.

Multi-Agent Workflow Design

Design agentic loops, orchestrate multi-agent systems, and manage task decomposition across workflows.

Tool Integration & MCP Configuration

Connect tools with clear boundaries, integrate MCP servers, and distribute capabilities across agents.

Setup & Configuration

Configure CLAUDE.md hierarchies, create custom skills, apply company-specific rules, and integrate into existing processes.

Quality Control & Structured Output

Design prompts with explicit criteria, enforce structured output formats, and implement validation loops.

Reliability & Context Management

Preserve critical information across long interactions, manage error handling, and ensure consistent results.

Example ZPT Folder Structure

Not every engagement includes a folder. But when it does, this is what it looks like.

example-zpt/

├── company-context/Context — What the business is: overview, product, ICP, positioning

├── account-management/CRM — Prospect data and profiles, organized by pipeline stage

├── skills/Standards — How output should look: templates, formats, quality rules

├── memory/Memory — Lessons learned, preferences, patterns discovered over time

├── output/Deliverables — Drafts, content, reports ready for review

├── workflows/Processes — How tasks get done: step-by-step procedures

├── mcp/Tools — Available integrations and tool configurations

├── CLAUDE.mdInstructions — Operating instructions for the AI agents

└── AGENT.mdAgent config — Agent-specific settings and preferences

The structure above is an example. Every deployment is customized to match the company's actual workflow, tools, and processes. 100% company-owned. No data leaves your network.

Screenshot from Claude Desktop, pointing at a ZPT folder

Claude finding new prospects by checking HubSpot and searching Google Maps

Real example: Claude finds new prospects in Chicago by checking HubSpot, cross-referencing the ICP, and searching Google Maps.

Connects to Your Stack

ZPT plugs into the tools your company already uses. No migration, no new software.