Self-Driving Finance

The trading app for
self-driving finance

Put your crypto on autopilot-starting with AI for Polymarket.

Trade on Autopilot Stats
autopilot-session_active
mode: self-driving // fully autonomous
route: momentum_reversal // compiled from NL thesis
lanes: polymarket, kalshi
scanning: 284 markets // orderbook + wallets + sentiment
signal: BTC_100K_JUN → smart money accumulating, spread compression
maneuver: BUY YES @ 0.62 $50 · confidence: 0.84
safety: ✓ all checks passed // budget, exposure, risk, session
status: executed // clientOrderId: 0x8f3…a1d
where are we driving next?
Momentum Agent bought YES on ETH_ATH_Q2 0x4a2…f8c
Contrarian Whale shorted TRUMP_WIN @ 0.54 0xd63…a1d
News Scalper exited FED_RATE with +12% PnL 0x7b1…e92
Whale Mirror copying top-10 wallet on BTC_150K 0xa8f…3dc
Arb Agent cross-venue spread on SOL_200 0x2c9…b47
Sim Runner backtesting mean_revert · 90d 0xf1e…704
Momentum Agent bought YES on ETH_ATH_Q2 0x4a2…f8c
Contrarian Whale shorted TRUMP_WIN @ 0.54 0xd63…a1d
News Scalper exited FED_RATE with +12% PnL 0x7b1…e92
Whale Mirror copying top-10 wallet on BTC_150K 0xa8f…3dc
Arb Agent cross-venue spread on SOL_200 0x2c9…b47
Sim Runner backtesting mean_revert · 90d 0xf1e…704
Features

Every lane of the market,
covered on autopilot

From idea to trade, every component runs independently-like a self-driving system with sensors, navigation, and safety built in.
Navigation
Tell it where to go, not how to drive
You wouldn't give a self-driving car turn-by-turn directions. Same here - describe your conviction in plain English, and the strategy compiler plots the route: entries, exits, sizing, guardrails.

Thesis-first design

No code, no config files. Speak your market view and the compiler maps it to executable logic.

Transparent routing

Every compiled rule is visible and editable. You always see the map before the car moves.

Fork & iterate

Branch strategies like code. Run v1 live while backtesting v2 on the side.

Your wallet, your vehicle

Agents operate on your existing wallet via scoped session keys. You hold the keys - literally.

strategy.compile("Buy YES on crypto markets
  where smart money is accumulating
  and sentiment shifts bullish")

// compiled output:
entry: smartMoneyFlow > 0.6
       && sentimentDelta > 0.15
exit: pnl > 12% || drawdown > 5%
risk: maxPosition $200, maxDaily $500
venue: polymarket, kalshi

// v2-forked from v1, wider entry
status: ready to deploy
autopilot.ai analyzing portfolio...

insight: Your momentum strategy is
  correlated 0.82 with BTC price action.
  Suggesting contrarian hedge:

proposed: SHORT YES on ETH_FLIP_BTC
         when momentum divergence > 2σ
         // reduces portfolio correlation 0.82→0.41

⚠ advisory only-requires your approval
Sensors
360° market awareness, always on
A self-driving car has LiDAR, cameras, and radar running simultaneously. Autopilot AI does the same for markets-scanning orderbooks, wallets, sentiment, and news in parallel to spot what you'd miss.

Signal discovery

Surfaces patterns and opportunities across hundreds of markets, matched to your thesis and risk profile.

Adaptive suggestions

Learns from your strategy's behavior and market shifts. Proposes adjustments-never overrides.

Correlation radar

Detects hidden exposure overlap across your strategies and suggests hedges before drawdown hits.

Advisory, not autonomous

Sensor layer informs decisions with confidence scores. It can shift conviction-it can't bypass your policy engine.

Safety Systems
Guardrails at every layer
Self-driving systems don't just accelerate-they know when to brake, when to yield, and when to pull over. Autopilot's policy engine enforces limits you set before a single dollar moves.

Dashboard view

PnL, exposure, drawdown, and Sharpe across all running strategies-one screen, real-time.

Configurable speed limits

Per-strategy budgets, per-order sizing, daily loss caps, position concentration bounds. You define the envelope.

Kill switch

Revoke session keys, pause any agent, or shut down all execution instantly. Control never leaves your hands.

Black box recorder

Every tick, every signal, every pass-logged with the full reasoning chain. The flight recorder for your portfolio.

portfolio.status

strategies: 3 active, 1 paused
total_budget: $2,400 remaining
realized_pnl: +$342.80
max_drawdown: -4.2% // limit: -8%
open_positions: 7

risk_alerts: 1 warning
  momentum_v2 approaching daily loss cap
  ($48 of $50 limit used)

permissions:
  session expires in 4h 12m
  allowed: polymarket/crypto/*
evidence_pack // tick #847, BTC_100K_JUN

market: spread 0.02, depth $14k, trend ↑
traders: smart money net +$8.2k (24h)
news: topic_intensity: 0.7, sentiment: +0.3
crypto: BTC $97.4k, vol_24h: moderate
sports: n/a for this market

quality:
  staleData: false
  missingSources: 0
  conflictScore: low

  → ready for decision
Road Data
Every signal feeding one decision
Self-driving needs HD maps, traffic data, and weather-all fused into one world model. Autopilot merges orderbook depth, wallet flows, news velocity, and social signal into a unified evidence pack per tick.

Market microstructure

Orderbook snapshots, trade flow, spread dynamics, and short-horizon price trends from Polymarket and Kalshi.

Wallet intelligence

Smart money tracking, position changes, copy-trade candidates, and leaderboard movement via Predexon.

Sentiment & event layer

Real-time topic intensity from X, news event detection from GDELT, and trending signal analysis. Advisory weight only-never overrides market data.

Bring your own signal

Plug proprietary data into the pipeline. Modular adapters mean any source becomes a lane on the road.

Architecture

Under the hood

Isolated systems, clear boundaries. Like a vehicle platform-swap any module without rewiring everything else.
Step 1
Strategy Compiler
Step 2
Tick Decider
Step 3
Policy Engine
Step 4
Execution
Data
Polymarket CLOB
Data
Predexon Wallets
Data
Sentiment / News
Data
CoinGecko / Odds
Infra
Convex Realtime DB
Auth
Privy + Session Keys
Observability
Action Timeline · Audit Log · Decision Trace · Simulation Metrics
How it works

From ignition
to destination

Five systems, one flow. You set the course-the rest is automatic.
1

Set destination

Write your market thesis in plain language. "I think crypto prediction markets are mispricing BTC momentum-buy YES when smart money accumulates and sentiment flips." The compiler maps it to executable rules.

2

Read the road

Each tick, the agent scans the environment-orderbooks, wallet flows, news events, sentiment-and builds a real-time evidence pack with quality flags on every source.

3

Decide the maneuver

The decision engine weighs the evidence against your strategy rules and proposes an action: buy, sell, or hold. Full reasoning chain attached.

4

Safety check

Before any order touches the market, the policy engine runs deterministic checks-budget limits, exposure caps, daily loss bounds, session validity. Fail any check → no trade, logged reason.

5

Execute & log

Approved orders are signed via your delegated session key and submitted. Every action-including passes-is recorded with tool calls, policy receipts, and external refs. Your black box.

Compare

Manual driving vs. autopilot

You can steer yourself, use cruise control, or let the car drive. Here's the difference.
Capability Autopilot Phantom / Polymarket Bankr
Natural language strategies
Multi-source intelligence 6+ live feeds Manual research Chat-based
AI-powered decision engine Per-tick LLM reasoning Conversational
Autonomous execution Fully autonomous Click-to-trade Assisted
Non-custodial
Decision audit trail Full reasoning chain Tx history only Chat logs
Simulation / backtesting Full sim engine
Risk management layer Deterministic policy engine Manual limits Basic
24/7 hands-free operation
Early Access

First passengers

From degens running single strategies to funds managing multi-market exposure.
"I used to spend 4 hours a day watching Polymarket orderbooks. Now my agent catches the same setups at 3am while I sleep. It's not even close."
- Early Access Trader
"The black box recorder is what sold me. I can replay any tick and see exactly what the agent saw, what it considered, and why it passed. That's institutional-grade transparency on a retail product."
- Fund Operator, Beta
Roadmap

The road ahead

Shipping fast. Building in public. Every milestone brings more lanes to the highway.
Q1 2026

Foundation

  • Polymarket agent execution
  • Strategy compiler (NL → logic)
  • Non-custodial wallet + sessions
  • Simulation engine v1
  • Observability & action timeline
  • Deterministic policy engine
Building now
Q2 2026

Multi-venue & intelligence

  • Kalshi + Hyperliquid integration
  • Cross-market arbitrage agents
  • Smart money + sentiment signals
  • Strategy marketplace (share/clone)
  • Data marketplace (BYOD)
  • Advanced sim & backtest metrics
Next
H2 2026

Scale & institutional

  • Multi-chain agent deployment
  • Agent-as-a-service layer
  • On-chain performance proofs
  • Institutional delegation & controls
  • Mobile monitoring app
  • Custom execution kernels
Planned
FAQ

Everything you need to know

What exactly is "self-driving finance"?
Same idea as a self-driving car: you set the destination (your thesis), define the rules of the road (risk limits), and an autonomous system handles navigation, obstacle avoidance, and execution. You can watch, intervene, or just check results later.
Does Autopilot hold my funds?
No. You connect your existing wallet and grant scoped session key permissions - think of it like valet parking with a speed limiter and a geo-fence. You revoke the keys anytime, and the agent can never exceed limits you set.
How does the AI actually decide to trade?
Your thesis gets compiled into structured rules. On every tick, the agent builds an evidence pack (orderbook, wallet flows, news, sentiment), runs it through an LLM decision engine, then passes the proposal through a deterministic policy engine. If any safety check fails, it's an automatic no-trade. Every step is logged.
Can I test drive before going live?
Yes. Simulation mode replays your strategy over historical data with configurable slippage, latency, and fee assumptions. Sim agents are fully isolated-they physically cannot place live orders. Think of it as the test track before the highway.
How is this different from Bankr or trading bots?
Bots follow static rules. Bankr gives you a chat copilot. Autopilot is a full autonomous system-it perceives (multi-source data), decides (LLM reasoning), checks (policy engine), and executes (signed orders) in a continuous loop. It's the difference between cruise control and actual self-driving.
What happens if something goes wrong?
Multiple layers: per-order limits, daily loss caps, session key expiry, and an instant kill switch. If critical market data goes stale, the agent auto-defaults to no-trade. You can pause or revoke any agent in one click. The black box recorder lets you replay exactly what happened and why.
Is it free?
Early access is free. We'll introduce tiers as multi-venue, advanced data, and institutional features roll out.

Stop driving manually.
Switch to autopilot

Self-driving finance for prediction markets. The road is open.

Trade on Autopilot Stats