solana-jupiter-arb

HFT-grade triangular arbitrage scanner for Solana, using Jupiter Aggregator quotes as the price oracle and Telegram for real-time opportunity alerts.

The scanner walks a configurable list of three-leg cycles (USDC → SOL → JUP → USDC and similar) and on every tick requests three sequential quotes from Jupiter for each cycle. Anything that returns more USDC than it started with — net of fees — is logged and, above the configured basis-point threshold, pushed to a Telegram chat for execution review.

Why self-hosted, not paid API

This is a latency-critical workload. The free Jupiter API tier is rate-limited; even the paid tier doesn't help where it actually matters:

Constraint	Free	Paid	Self-hosted Metis
Rate limits	strict	relaxed	none
Network RTT	~50ms	~50ms	~1ms (loopback)
Determinism	upstream-load dependent	upstream-load dependent	local
Custom routing	❌	❌	✅

A single triangle is 3 sequential quotes. At 50ms RTT each, a single triangle scan is 150ms — half a Solana block (block time ~400ms). Arb windows on Solana close in 1-3 blocks. Self-hosted brings per-leg latency to single-digit ms via loopback, turning a 150ms triangle scan into a ~10ms one. That's the difference between catching an opportunity and watching it close.

Beyond raw latency:

1. No rate limit — a useful scanner runs 30-50 triangles continuously. At 1s cadence that's 90-150 req/s. No public tier sustains that without per-request cost.
2. Determinism — public API responses fluctuate with upstream load. Self-hosted gives consistent timing for evaluating routing decisions and backtesting.
3. Custom routing — the binary lets you set DEX inclusion rules and excluded pools. Relevant for any builder also running LP positions on routed pools.

Stack

• Node.js 20+
• Native fetch (no SDK dependency, no supply-chain surface)
• BigInt arithmetic for token amounts
• Telegram Bot API for alerts (no SDK, single HTTPS POST)
• JSON config for triangles, tokens, thresholds

Layout

src/
  index.js     — main scan loop with per-batch latency log
  scan.js      — three-leg cycle evaluator (returns bps profit)
  jupiter.js   — Jupiter quote client (public ↔ self-hosted toggle)
  triangles.js — cycle definitions
  tokens.js    — token registry (mint, decimals, symbol)
  telegram.js  — alert sender (rate-limited, no-op if not configured)
.env.example   — API URLs, Telegram credentials, thresholds

Status

Early. Triangle scan loop runs against the public API and prints profit/loss in basis points per cycle. Telegram alerts fire above the configured threshold. The scanner does not execute swaps yet — opportunity detection only.

Roadmap

• Triangular arb scanner (3-leg cycles, USDC base)
• Per-leg latency measurement
• Configurable basis-point alert threshold
• Telegram notifications with per-cycle rate limiting
• Concurrent triangle scanning with rate-limit budget
• N-leg cycle support (4-leg, 5-leg cycles)
• Multi-DEX direct quotes (Raydium, Orca, Meteora) for cross-DEX arb in addition to triangular
• Yellowstone gRPC stream subscription for on-chain pool state (event-driven scans, not just polling)
• Jito bundle submission for atomic execution of top opportunities
• Self-hosted Metis binary integration (drop-in URL swap via env)

Run

cp .env.example .env
# fill in TELEGRAM_BOT_TOKEN and TELEGRAM_CHAT_ID for alerts
npm install
npm start

Flip USE_SELF_HOSTED=1 once the Metis binary is running locally on the URL in JUPITER_SELF_HOSTED_URL.