The brainstem’s system prompt mentions three tiers:

The user may be at any stage of the RAPP journey:

  • Tier 1 — Brainstem: local Python or this browser-only twin (where they are now)
  • Tier 2 — Hippocampus: Azure Functions with persistent memory, runs locally first
  • Tier 3 — Nervous System: Copilot Studio + M365/Teams

The tiers are a deliberate ladder, not a feature matrix. Each one builds on the previous. You climb when you need to, not before.

Tier 1 — Brainstem. This is where everyone starts. A local-first AI assistant that runs on your machine (or in your browser, via the virtual brainstem twin). It chats. It calls agents. It remembers things to localStorage or a JSON file. It does not need Azure, does not need API keys, does not need a backend. The brain metaphor: this is the brainstem — core reflexes, the part of the brain that runs whether you’re paying attention or not.

For most users, Tier 1 is the whole product. Most workflows are: chat with my AI, have it call a few agents, get an answer, move on. You don’t need cloud infrastructure for that. You don’t need persistent multi-user memory. You don’t need to expose your AI in Teams. You need a local AI that can do work.

Tier 2 — Hippocampus. When you’re ready for memory that survives more than a single machine — when you want your AI to remember things across devices, share state with teammates, or access files in cloud storage — you graduate to Tier 2. This is Azure Functions with persistent memory. The brain metaphor: the hippocampus is where short-term memory consolidates into long-term memory.

You install Tier 2 with one line. The brainstem’s soul prompt knows this and offers it when users ask for “the cloud” or “Azure” or “memory that persists”:

curl -fsSL https://raw.githubusercontent.com/kody-w/rapp-installer/main/community_rapp/install.sh | bash

It scaffolds a project at ~/rapp-projects/{name}/ with its own venv, runs locally first, can deploy to Azure later when you actually want shared state.

Tier 3 — Nervous System. When you need your AI to be reachable from inside your organization — Microsoft Teams, M365, Copilot Studio agents — you go to Tier 3. The brain metaphor: the nervous system reaches out from the brain to the rest of the body.

This tier is heavyweight. It’s where you stop being a single user with a personal AI and start being an organization that’s deploying AI to its workforce. There’s identity, there’s RBAC, there’s compliance, there’s deployment pipelines. The brainstem helps you get here, but it doesn’t pretend the climb is one curl command. Tier 3 is real engineering work.

The point of the ladder is that each tier is whole on its own. You can stop at Tier 1 forever. Many users will. The brainstem doesn’t push you up. The system prompt explicitly says: don’t push users to Azure or Copilot Studio — let them ask when they’re ready.

This matters because most “AI platform” products try to sell you the cloud version on day one. RAPP’s bet is the opposite: start local, prove value, climb when you have a reason. Most personal AI users never need to leave the brainstem. The ones who do, do because their use case actually requires it — multi-device, multi-user, enterprise reach. Not because we made them.

The ladder is also the upgrade path. Your Tier 1 agents work in Tier 2 unchanged (same __manifest__, same perform()). Your Tier 2 agents are reachable from Tier 3 via Copilot Studio’s agent action surface. You don’t rewrite as you climb. You just add.