PayCargo · AI Platform · TLDR · 3 of 5

The Cartridges

The actual rides. Each one a piece of equipment kids can climb on.

A cartridge is an AI app that has been mounted on the playground. AI Automation is a slide. The observability dashboard is a seesaw the operator rides. The future code-agents app will be the monkey bars. Each cartridge is self-contained but inherits every safety feature for free.

The naming matters. We do not call them "apps" because that implies they each handle their own gate, their own bouncy floor, their own cameras. They do not. We call them cartridges (like an NES cartridge) because the playground is the console, the cartridges plug in, and the cartridge gets all of the safety stuff from the slot it plugs into.

What rides we have today

Observability Live

The operator's clipboard. Built first because you cannot run a platform you cannot see.

The screen where the operator looks at health, cost, audit logs, scheduled jobs, users and groups, and per-cartridge spending. Also holds the documentation tab (this one).

Lives at cartridges/administrative/obs/. Operator-facing only.

AI Automation (ap-invoice) Live

Freight document automation — the first customer-facing ride.

Ingest freight documents, classify them with Bedrock, extract the fields, match vendors and post. It is tenant-aware: it follows the tenant picked in the Console top bar and its backend re-checks it on every call (see How tenants keep customer data apart). Vendor matching and posting run through the decision core, a shared service with its own Postgres database and per-tenant row-level security.

Lives at cartridges/finance/ap-invoice/.

Mock TMS Live

A TMS financials simulator to test against.

An accrual ledger, invoice matching, “Sound Board” variance levers and reset-to-golden — a stand-in for a customer's transport management system so AI Automation can be exercised end to end without a real one.

Lives at cartridges/finance/mock-tms/.

Mock Email Live

A mail sink: see exactly what an app would have sent.

Point an app here instead of a mail provider and read the headers, both bodies, every attachment, and what is wrong with the message. It cannot deliver anything, by design. (Real system mail is sent separately, as no-reply@paycargo.ai.)

Lives at cartridges/global/mock-email/.

Backlog Live

One queue across every cartridge, the Console and the foundation.

Projects, epics and intake, worked from Claude Code over MCP; feedback sent from the Console top bar lands here. Active time is inferred from events rather than a timer.

Lives at cartridges/ai-test-kitchen/backlog/.

Doc Vault Live

A versioned, shareable home for the HTML documents the platform produces.

Projects of versioned files; every file versions immutably; documents render as live artifacts behind a security boundary, are shared to named people, and carry comments pinned to the exact version you read. It's also where automated writers publish. See Step 6.u.

Lives at cartridges/global/doc-vault/.

Foundation Audit Live

The platform audits itself, on a schedule.

Read-only AI audits (discipline drift, SOC 2, cartridge quality) run on EventBridge schedules and publish reports into Doc Vault. Its console lets an operator run one now, watch run history, edit the doctrine the auditor follows, and connect an external scheduled evaluator. See Step 6.w.

Lives at cartridges/administrative/foundation-audit/.

Not in this environment

Earlier versions of this page listed Chatbot, SquadFlow, Sales Reporting, Pipeline Analysis and Integration Workbench as live. They ran in the retired account and are not in the PayCargoDevOps/AI-Automation-Shared repo, so they are not deployed here. The rides above are the eight apps live in account 753974371195, counting the Console and obs.

What rides are on the roadmap

Marketing, HR, Finance Folders reserved

Each department gets its own folder. Department-led, platform-supported.

These department folders exist with placeholder markers (cartridges/marketing/, cartridges/hr/, cartridges/finance/). As a department identifies its first AI use case, its first cartridge lands there. (Labs is no longer empty — it holds Integration Workbench.)

Not rides, but part of the platform: some capabilities live in the Console itself rather than as a cartridge — the Content CMS (versioned CLAUDE.md/skills.md that steer AI calls) and the Console MCP (machine tokens that let a scheduled Claude Code read doctrine and publish reviews). Cartridges and scheduled evaluators use them; operators manage them from the Console.

What makes a cartridge a cartridge

Specific shape, intentional. Every cartridge is:

PropertyWhat it means
Self-containedOne folder under cartridges/<department>/<name>/. Has its own Terraform, frontend, Lambda code, docs.
Plug-and-playThe Console auto-detects it. No separate registration step. New cartridge folder + apply equals a new tile on the launcher.
Foundation-inheritedEncryption, identity, audit, supervisor, network limits all apply without any cartridge code. The cartridge author writes business logic.
Standardized backendAPI Gateway HTTP API plus Lambda. JWT-verified. Origin-secret protected. CloudFront-fronted.
Standardized frontendS3 plus CloudFront. Loads the SDK from the Console. Uses PayCargo design tokens.
Operator-visibleShows up on the obs Cartridges tab automatically. Cost attribution, last-invoked timestamps, configuration panel all free.

How a new cartridge gets born

  1. The Anvil sketches it. Name, capabilities, OAuth providers needed, data classification, cost estimate. (See Anvil.)
  2. The Forge builds it. AI-assisted scaffolding from the spec. (See Forge.)
  3. The Console mounts it. A pull request lands, an apply runs, a new tile appears on the launcher.
  4. The Foundation protects it. Without any cartridge author work, the safety features apply to every call the cartridge makes.

Cartridges per department, not per app

The folder structure says cartridges/sales/, cartridges/marketing/, etc. on purpose. A department can have many cartridges over time. The folder is the home. The first one lands; the second, third, fourth land next to it. Department-led AI strategy is structurally encouraged.

Want the technical version?

Cartridge engineering depth lives in Step 6.d (repo structure), Step 6.i (cartridge metadata), Step 6.j (chatbot build walkthrough), Step 6.l (code-agents spec), and Step 4.5 (observability cartridge).