The Model Broker

For an AI team lead paying several model providers.

One signal · 7 specialists · one result

Speaks to people1
Keeps knowledge2
Runs infrastructure2
Audits and guards2

How the work moves

Four steps, each done by one specialist

Starts with

A model call from any agent

First

1Sends each call to an approved provider within its rate limitAI Provider Switchboard 2Registers commercial assistants and holds each grant for a personAI Access Desk

Then

3Charges every call to the task that made itAI Spend Recorder 4Turns the totals into cost per projectEngineering Cost Analyst

Ends with

Model spend charged to each task and project

the signalthe resultEach step is done by one specialist on the crew

The story

Every model call from every agent passes through one switchboard. The agent sends each call to an approved provider, keeps each provider inside its rate limit and steps down to a cheaper tier when a quota runs out. Commercial AI assistants that want access are registered, and a person approves each grant. Every call is charged to the task that made it.

The crew

The specialists behind this role, grouped by the kind of work they do.

Audits and guards

AI Access DeskRegisters each AI assistant as an app, holds every grant until a person approves it, accepts sign-in tokens from your own server and from your workplace platform, shows each assistant only the tools it is allowed, and ties every action to the assignment rather than to what the assistant claims.Also serves The Security Officer, The Assistant-Ready Website
Engineering Cost AnalystAttributes every model call to the task, role and phase that spent it and turns the totals into cost and human-equivalent effort hours per project.Also serves The Spend Controller

Runs infrastructure

AI Provider SwitchboardSends every agent's model calls to an approved provider, keeps each provider inside its rate limit, steps down to a cheaper tier when a quota runs out, and blocks traffic that a provider's terms do not allow.Also serves The Spend Controller
Model Comparison EvaluatorThis agent puts two or three model versions through the same hidden test set, runs any generated code safely, scores the answers strictly and explains where each version fails.Also serves The Model Launch Manager

Keeps knowledge

Memory Return-on-Investment ReporterReports the tokens spent creating each note against the tokens saved each time a note is reused or compressed, priced per model provider on per-token or flat-fee plans, and shows project-level totals on a live dashboard.Also serves The Search Auditor
AI Spend RecorderIt reads the model usage from the session, removes duplicates, and records the cost against the task it belongs to.Also serves The AI Workforce Manager, The Spend Controller

Speaks to people

Tool ConciergeTakes a plain-language need and returns the best-matching tools from a catalog of hundreds in one answer.Also serves The Team Architect

Where this crew also works

The Spend Controller33

Each bar counts the specialists this role shares with another.

The engineering behind this role

AI subscriptions as governed workersCapability Role contracts, with vendors as dataIdea What the work cost, to the pennyEngineering story All capabilitiesThe engineering, by what it does