The 11 agents

A run, end to end

This is one delivery, drawn on a time axis. Each bar is an agent doing its single job, and the whole thing plays out left to right: intake becomes a plan, the plan becomes infrastructure, and the infrastructure moves toward a signed-off handover. Press play, or drag the timeline to move through the run and watch each stage settle.

Bars in the brand color are agents doing reasoning work; the lighter bars are the deterministic engines and integrations they hand off to. The Compliance check retries until it passes, Zoho is served from cache when the project already exists, and everything downstream stays queued until its input is ready. Read it once and the rest of this page is just naming what you already saw move.

The orchestrator

Every agent is a specialist that does exactly one thing. What turns eleven specialists into a delivery pipeline is the orchestrator: a LangGraph state machine that sequences the agents, checkpoints run state, pauses at the human approval gates, and streams progress to the run canvas. The orchestrator owns the order of work and the shared state, so each agent can stay focused on its own task without knowing about the others.

The orchestrator treats a run as a sequence of nodes with saved state at each step. If a critical node fails, the run halts rather than pushing forward on bad output, and it can be retried without re-running the steps that already completed. Non-critical nodes such as IPM are allowed to warn and continue.

Agent groups

The eleven agents cluster into five stages that mirror the shape of a delivery. Reading them as groups is the fastest way to understand what the platform does before looking at any single agent.

Initiation

The initiation group prepares the ground before any analysis runs. IPM sets up scoped, temporary access into the target AWS account, and the Zoho Agent creates the delivery project shell. This group turns an approved request into a working environment with credentials and a project to track against.

Analysis

The analysis group turns the document into a plan. The SOW Analyzer reads the SOW, extracts structured requirements, and builds the infrastructure query and work breakdown. The Routing Engine then decides, deterministically, which services are handled by templates and which need generated Terraform. The output is a concrete plan of what to build and how.

Generation

The generation group produces the infrastructure code. OneClick provisions the base, template-covered services at no AI cost, while TerraFormGen writes everything else as additional Terraform. A Compliance validator checks the combined output against the Workmates rule set and applies safe auto-fixes, so what reaches review is already consistent and tagged.

Deploy

The deploy group moves reviewed code into a running system. The GitHub Agent opens a pull request that carries the plan and a CI workflow, and the Ansible Agent configures the instances after they are provisioned. Nothing in this group applies changes until a Tech Lead has approved at the gate.

Handover

The handover group closes the loop with the customer. The Documentation agent assembles a branded runbook, and the Signoff agent produces the signoff document and routes it for dual signature before the project is closed. Both agents are planned and not yet deployed, so this group is documented ahead of general availability.

Agent reference

The table below lists all eleven agents and the single job each one owns. Agents marked as planned are designed but not yet deployed.

Agent	Responsibility
SOW Analyzer	Parses the uploaded SOW, extracts structured requirements, and builds the infrastructure query plus the work breakdown structure, checklist, and questionnaire. Reuses learning from past SOWs.
TerraFormGen	Generates, modifies, and self-validates the per-service Terraform using the standard Workmates CWM module library, retrying validation until it passes.
Platform Backend	The API router that the other services call through, and the surface that runs the deterministic service-routing decision.
Zoho Agent	Creates the delivery project, syncs the work breakdown as milestones, tasklists, and tasks, advances the deal stage, and drives digital signatures for signoff.
Ansible Agent	Runs post-deployment configuration playbooks on the provisioned instances through AWS SSM, with no direct SSH access.
IPM	Identity and Privilege Management. Sets up scoped, temporary cross-account AWS access before analysis. Non-critical: the pipeline warns and continues if it fails.
OneClick	Template engine that provisions base infrastructure such as VPC, EC2, RDS, ALB, S3, and KMS at zero AI cost, and doubles as a standalone provisioning catalog.
GitHub Agent	Creates the repository, pushes the Terraform and modules, generates the CI workflow, and opens a pull request with the plan attached. Connects to AWS with keyless OIDC.
Audit	Planned lightweight security scan over the generated Terraform. Today it is a transparent pass-through so the pipeline continues; deploy-time gating is handled by the CI security scan and the Compliance check.
Documentation	Planned runbook generator that will extract resources from Terraform state, capture console screenshots, and assemble a branded runbook. Designed, not yet deployed.
Signoff	Planned agent that builds the branded signoff document, routes dual digital signatures, and stores the signed artifact for handover.