Rescue Team AI

Gamified multi-agent AI orchestration — Pokémon Mystery Dungeon style

Work in Progress

The Big Idea

Managing multi-agent AI workflows is highly technical and usually about as visually exciting as reading server logs. Rescue Team AI fixes that by wrapping agent orchestration in a Pokémon Mystery Dungeon visual theme.

You build a "rescue team" of specialized AI agents, each stationed at a shop on a 2D market map. An orchestrator agent (your team leader) physically walks across the map to delegate tasks to sub-agents — all driven by your prompts in a single chat window. It's agentic AI you can actually watch.

Pokémon nerd not required for use. But it helps.

How It Works

Supervisor Routing: You chat exclusively with the team leader. It decides which specialist to consult.
Market Architecture: Agents live at shops on a 2D canvas map. The leader sprite walks to them in real time.
Local Execution: Agents interact with your local file system — cloned repos, terminal commands, the works.
State Polling: HTMX-driven status updates show when an agent is thinking and when a response is ready.

The Stack

AI Engine Python + Pydantic AI

LLMs Anthropic · OpenAI · Gemini

Backend Django + SQLite

State HTMX async polling

Game Engine TypeScript + 2D Canvas

Hosting Digital Ocean Droplet

Roadmap

Month 1

Backend infrastructure — SQLite, Pydantic AI orchestration, basic local git tools.

Month 2

Frontend foundations — Django server, HTMX polling, static map rendering, landing page deploy.

Month 3

Visual logic — TypeScript sprite animations and market routing behavior.

Month 4

MVP polish — Refine the chat interface, bug testing, and launch.

Follow the Build

This project is actively in development. Star the repo, watch for updates, or clone it and poke around.

View on GitHub →