From research project to a non-custodial trading toolkit.
Arblane started as a personal research project — 90 days of recording Polymarket ticks, scoring resolution risk, and hand-running basket arbitrage in a notebook. The patterns kept repeating; the tools didn't exist. So we built them.
The product is what we'd want as users: paper-first, custody-free, with measurement everywhere. The platform earns subscription fees. Trading P&L stays with the trader, including losses.
What we won't compromise on.
01
Non-custodial first
We never hold user wallet keys. Not in the database, not in Secrets Manager, not in memory. Every live order is built as EIP-712 typed-data, signed in your wallet, then relayed. The platform is structurally incapable of moving your funds.
02
AI in research, not execution
Strategies are deterministic functions. AI helps with research, monitoring, and post-trade analysis — never sits in the decision path. The one exception is the Phase 5 news-reaction strategy, and that runs under strict confidence + size + daily-loss-limit guards, with the user signing every trade.
03
Measurement before scale
No marketing claim ships without a measurement behind it. Drift, P&L attribution, decision deep-dive, resolution-risk scoring — these came first. Then strategies. Then features. Then this site.
AWS-only. Node 22. TypeScript everywhere.
Node 22 LTS · TypeScript strict
Next.js 16 · MUI v9 · App Router
Fargate · Prisma · Postgres · Valkey
S3 · Parquet · CLOB V2 WS
Better Auth · Passkeys · TOTP
Reown AppKit · WalletConnect · viem · SIWE
Stripe (provider-abstract layer)
AWS CDK · eu-west-1 only
Founder · Arblane
Solo-built so far. Background in distributed systems and quant tooling. Reach me at hello@arblane.com.