Simple

RAPP Agent Repository

Browse and deploy single-file AI agents for your business.

/

Loading registry...

Curator

Human

Card Rankings

Gauntlet

Curated by our virtual artist in residence — take with a grain of salt


Loading...
Every agent tells a story

AGENT CARDS

Inspired by @borg/cardsmith_agent by Howard — the original CardSmith

One file, one purpose
A thousand agents waiting
Which deck will you build?

0 cards
?
Companion
None
All Official Verified Community Frontier | My Deck

Loading cards...
Build • Validate • Preview

AGENT WORKBENCH

Agents are generated by LearnNewAgent — describe what you want and it builds the agent.py for you. Use this workbench to inspect, validate, and preview before publishing.

A friendly name for your agent. This is what people see in the store.
What does this agent do? Keep it to one sentence.
Your agent publishes under @username/
Keywords that help people find your agent. Press Enter after each one.
Start with Frontier if you're still iterating. Community when it's solid.
Semantic versioning: MAJOR.MINOR.PATCH
API keys or config your agent needs at runtime. Leave empty if none.
@rapp/basic-agent
Other agents this depends on. @rapp/basic-agent is always included.
This becomes the docstring — the agent's built-in documentation.
What does your agent need from the user to do its job?
If you know Python, write the logic here. If not, leave it blank and we'll generate a starter.

Your Agent Card

Card will appear when you build.

Generated Code

Submit Your Agent

Build your agent with any LLM — ChatGPT, Copilot, Claude, Gemini — then drop the .py file here. We handle the rest.

1. Start from a template

Download a template, open it with your favorite AI, and tell it what you want your agent to do.

Blank Agent
Minimal skeleton
Hello World
Returns a greeting
API Caller
Calls an external API
Data Processor
Transforms input data
Click a template to download the .py file. Give it to any LLM with your instructions.
2. Drop your finished agent
+
Drop your _agent.py here
or click to browse

Your agent goes to staging for review. Once approved, the forge mints your card and you're in the registry.

Collect • Grade • Trade

BOOSTER PACKS

Open packs to mint unique, graded agent cards. Every card has a serial number, visual variations, and a condition score that changes over time.

Available Packs

Your Collection (0 cards)

No cards minted yet

Open a pack above to start your collection.

Trophy Case (showcase up to 6 cards)

Pull History

No pulls yet — open a pack!
Agent Challenge

Agent Quiz

Test your knowledge! Identify 5 agents from your collection.

Stream • Hot-Load • Steer

AGENT STREAMING

Stream agent bytecode from GitHub into memory. No download. No disk write. Hot-load, steer, eject. Close the tab and it's gone.

Unencrypted
loading... 3S

Runtime Targets

B
RAPP Brainstem
Disconnected
localhost:8200
On-device AI runtime. Streamed agents route to local perform() loop.
C
CommunityRAPP
Disconnected
—
Cloud/hosted runtime. Agents execute in managed environment.
Lp
Lisppy Runtime
Click to stream
github.com/kody-w/lisppy — streamed from raw
Scheme-like Lisp interpreter for agent systems. Streams from GitHub raw into this tab. Zero install. Data sloshing REPL: Read state → Eval agents → Print mutations → Loop.
O
Ollama
Click to configure
localhost:11434
True local-first AI. Run Gemma 4 on-device — no cloud, no API key, no data leaves your machine. Brainstem reasons with local models.
W
Store (Inspect)
Always On
Browse, stream, card, mint, export — no execution

How it works

Stream — Fetches raw .py from GitHub into a JS Map. No file write. Purely in-memory.
Hot-Load — Manifest is parsed, agent appears in browse/cards/search instantly.
Deck Router — Stream an entire deck. Each agent gets a GUID. The deck topology defines the routing group.
Steer — Route state through GUID-mapped agents. The same loop that fires perform() can be steered by swapping which agents are live.
Eject — Remove from memory. Gone. No trace on disk.
Brainstem — Connect to a local RAPP Brainstem. Push agent bytecode for on-device execution. The browser steers, the Brainstem performs.
Ollama — True local-first AI. Run Gemma 4 on-device via Ollama. The Brainstem reasons with local models — no cloud, no API key. ollama launch brainstem --model gemma4
CommunityRAPP — Route to a hosted instance. Same agents, cloud execution surface.
Flick — Pick a card, attach a task, flick it to a runtime. The card carries bytecode, the flick carries intent. Multiple flicks in parallel = multiple agents with their own data streams.
Encrypt — E2E encryption via AES-256-GCM. Generate a key, share via QR code on the card back. From scan to stream to steer — encrypted end-to-end.
Universal — The .py file is the unit. The runtime is a choice. Stream once, route anywhere.
RAPP Card Chain • Live Market

CARD EXCHANGE

Public chain. Every forged card is verifiable, transferable, and permanently on-chain.

Agent Cart (0)

Your cart is empty. Browse agents and click + to add them.