Architecture

The engine

Every run is driven by a single engine. The orchestrator sits at the center like a chip on a board: requests arrive, it sequences the specialized agents, and structured data flows back out to each one along its own path. The rest of this page unpacks that picture, from the layers around the orchestrator to the individual services it calls.

The layers

The platform is built in four layers. The Frontend starts runs and handles approvals. It talks to the Platform Backend, an API router that fronts the orchestrator. The LangGraph orchestrator sequences the specialized agents, and the agents do the delivery work. Each layer speaks only to its neighbor, so the frontend never calls agents directly.

The orchestrator

The orchestrator is a state machine. It sequences the agents, checkpoints run state so a paused run can resume, and stops at every human gate. It streams progress to the run canvas as it advances. If a critical node fails, the run halts and can be retried without re-running the steps that already completed.

Services overview

Each service owns a single responsibility. The orchestrator calls them in the order the pipeline requires.

Service	Role
Frontend	Web app that starts runs, visualizes the pipeline, and handles approvals.
Platform Backend	API router that fronts the orchestrator and the agent services.
LangGraph	Orchestrator state machine that sequences agents and checkpoints run state.
SOW Analyzer	Extracts structured requirements and builds the infrastructure query.
TerraFormGen	Generates and self-validates per-service Terraform from the Workmates modules.
OneClick	Template engine that provisions base infrastructure with zero AI cost.
GitHub	Creates the repo, pushes Terraform, and opens the CI pull request.
Zoho	Creates the delivery project and drives digital signatures.
Ansible	Runs post-deployment configuration playbooks over AWS SSM.

Datastores

State is split across purpose-built stores, described here by role rather than by product or connection detail.