rmi-backend/ARCHITECTURE.md

RMI Backend — Architecture

Status: canonical Last updated: 2026-07-03

Overview

FastAPI monolith serving crypto risk intelligence via HTTP, Telegram, and MCP.

High-level flow

Client (Web / Telegram / AI Agent)
    ↓
nginx → rmi-backend container (:8000)
    ↓
FastAPI routers
    ↓
Services / scanners / DataBus providers
    ↓
Postgres / Redis / Qdrant / Neo4j / ClickHouse

Components

Path	Purpose
app/routers/	HTTP route handlers (thin)
app/scanners/	Token risk scanners
app/databus/	30+ third-party data providers
app/telegram_bot/	Telegram bot command handlers
app/mcp/	MCP manifest and tool manager
app/facilitators/	x402 payment verification
app/wallet_memory/	Wallet label + clustering subsystem

Storage

Store	Use	Container / Host
Postgres	Primary OLTP, Ponder indexer data	rmi-postgres
Redis	Cache, Celery queue	rmi-redis
Qdrant	Vector embeddings	rmi-qdrant (down)
Neo4j	Wallet graph / labels	rmi-neo4j
ClickHouse	Analytics, 39M address labels	rmi-clickhouse

Interfaces

Endpoint	Purpose
GET /health	Health check (currently unhealthy due to Redis)
GET /docs	Swagger/OpenAPI docs
POST /scanner/scan	Live token scan
POST /mcp/...	MCP tool endpoints

Current issues

• x402_tools.py (212 KB) and x402_enforcement.py (108 KB) violate the 500-line limit and should become separate services or split by domain.
• sqlite3 is used in 5 files for local state; should migrate to Postgres/aiosqlite.
• F-string logs and print() statements prevent structured log aggregation.

Deployment

ssh netcup
cd /root/backend/
docker compose pull
docker compose up -d rmi-backend