Platform Architecture

How Momo is built - from the Telegram bot to the multi-agent backend and user isolation model.

Overview

Momo is a multi-layer AI platform. Each user gets a fully isolated workspace with persistent memory, their own agent sessions, and private encrypted storage. Nothing is shared between users.

Telegram / Web App
↓ HTTPS webhook / REST API
Momo Backend (Node.js, Process Manager, secure port)
↓ Auth + routing
Per-user Agent (OpenClaw)
↓ Model selection
Claude Opus / Sonnet / Haiku
↓ Tools
App Builder · Yield Radar · Wallet Scorer · Memory · Web Search

Infrastructure

ComponentDetails
ServerVPS at [secure server] - nginx reverse proxy
BackendNode.js + Express, Process Manager managed, secure port
Domainmomo-agent.xyz - SSL via Let's Encrypt
Web Appapp.momo-agent.xyz - React + Vite
Docsdocs.momo-agent.xyz - static HTML
Bot@MomoAgentXYZ_bot - Telegram webhook

AI Models

RoleModelUse Case
App Builderclaude-opus-4-5Building full web apps (32k tokens)
Crypto Intelligenceclaude-opus-4-5Wallet scoring (RAWR), yield radar, money markets, liquidation monitor, smart money tracking, SQD real-time event streams
Agent Sessionsclaude-sonnet-4-6Persistent OpenClaw agent sessions
Fast Chatclaude-haiku-4-5Quick responses, routing decisions

User Isolation

Every user gets:

Crypto Stack

FeatureData Source
Solana wallet scoringOn-chain transaction indexer - parsed swaps, all DEXes
EVM wallet scoringReal-time chain indexer - Base + ETH logs/transactions
Yield dataDeFiLlama pools API - APY, TVL, 30d history
On-chain flow confirmationReal-time chain indexer - 25 chains
Hyperliquid scoringReal-time fills dataset
DEX flow radarReal-time swap event indexer - Base + ETH
Token metadataOn-chain asset metadata resolver

Security

Process Manager Processes

ProcessServiceRole
main-backend[secure port]Main backend - chat, builds, crypto, webhooks
momo-alerts-Alert engine - polls wallets, fires Telegram alerts
openclaw-simple[internal port]OpenClaw gateway - persistent agent sessions