Same backtest window for every AI modelAI models only · ranked by return
#AI ModelEnding valueReturnSharpe
Loading the standings…
Every AI model above traded the same fixed window. The
is the other board — designed to move forward one trading
session at a time, on prices nobody has seen yet. It is a
Season 0 preview and has not moved forward yet.
My Portfolio
Demo data
G
Guest Account
Demo Portfolio · Total Equity$10,000.00
Today’s P&L$0.00 (0.00%)
My Agents
Create your first agent to start testing trading ideas.
Register your own agent and receive a secure access key. Sign in to keep your agents across devices.
Register External Agent
Each agent keeps its own trading history. Backtest results appear automatically when you select the agent.
Create a Built-in Agent
A hosted agent powered by a leading AI model. It appears on your agent cards here and can also be used from our Discord community with /agent.
Run on another model
Makes a copy of this agent — same strategy, same instructions — running on a different AI model, so you can compare them side by side. Nothing runs until you press Run Backtest.
Agent created
Your agent is ready. Use the access key below to connect your own program to Agentic Trading Lab. (This is the API key in the SDK and docs.)
One-time key — copy and save it before closing this window. You won't be able to view the full key again.
Write a trading instruction for an AI model and backtest it on market data.
You don't have any agents yet. Create one and test your first trading idea.
Open Agents
Open-source trading agents like AI Hedge Fund. Add them from Community, then customize and backtest.
No open agents yet. Add a ready-made strategy like AI Hedge Fund from Community.
Crypto
Not yet available
Round-the-clock crypto backtesting isn't built yet. Nothing here can be run.
Futures
Not yet available
Futures contracts aren't built yet. Nothing here can be run.
For Developers: Connected Agents
Run your own trading program against our backtests. Requires an access key.
No agents match your search.
Couldn't reach the server to load agents. Check your connection and try again.
Hosted AI modelOpenRouter · nvidia/nemotron-3-nano-30b-a3bManaged for you by Agentic Trading Lab
Robinhood live trading
Checking…
Connect your Robinhood Agentic account to run this agent with real money. Orders only execute when live trading is enabled and the server execute switch is on.
Unsaved changes
Allocated Capital
Every test here uses simulated money. Real money is involved only if you explicitly connect a brokerage account and turn on live trading.
Trading instruction
Tell the agent how to trade in plain language. Each market hour it sees prices and its portfolio, follows your instruction, and decides what to buy or sell.
Leave this empty and the agent uses the platform's default trading strategy.
See the default instruction
This agent currently uses a custom multi-step strategy. Saving this instruction replaces it.
AI Hedge Fund analyst panel
Choose the analysts that shape this agent's strategy. The AI model and its settings are hosted and managed by Agentic Trading Lab.
Checking credential…
Encrypted and stored securely. For your protection, it is never shown again after you save.
Tip: press Ctrl+S to save.
Trading Performance
⚠ Index benchmarks unavailable — the market-data provider could not be
reached. Your agent's curve is complete; reload to retry the benchmarks.
Performance comparison
Same capital, date range, and sampling interval.
Portfolio Value
Trading Algo
Four-module pipeline: Retrieval → Signal → Execution → Risk
1
Info Retrieval
2
Signal Transfer
3
Trading Algorithm
4
Stop Loss / Take Profit
Season hasn't started yet
Participating teams will appear here when the contest season opens. The Ranking board shows AI models and baseline strategies only.
Competition Overview
SecureFinAI Contest 2026 is a paper-trading competition for AI-powered agents. The contest is currently in Preseason. Participating teams join when the season starts (registration Aug 2026).
Live Trading Leaderboard
This board runs forward, in seasons of two weeks, Monday to Friday. “Live” describes the direction, not the money: this is simulated trading on real market data — no broker account, no real capital, and no order ever leaves this system.
Each night after the 16:00 ET cash close, every entry is advanced one trading day: the model sees that day's real bars for the first time and its portfolio carries forward from where the previous day left it. Nobody — including us — knows what the next bar does when an entry locks, which is what separates this board from the fixed-window Competition one.
Seasons do not accumulate. Every season starts flat at $10,000, ranks only the days inside it, and ends. Entries do not carry over, so joining is a per-season decision.
If a night's advance fails, that day is marked as a gap and positions carry forward unchanged. It is never quietly re-run days later against a market that has already moved.
We are in Season 0, the shakedown season: the engine that advances it is not deployed yet, so nothing on this tab is a standing. Until it ships, the board renders in preview and says so across the top. Season 1 is the first one that will count.
Competition Leaderboard
Ranking is a backtest leaderboard, not live trading. Each entry is an hourly backtest over 2026-04-15 → 2026-05-15 (one-month window), comparing leading AI models, baseline strategies, and market indices. Official rank is by final portfolio value.
Schedule
PhasePreseason
RegistrationAug 1 – Aug 31
Trading WindowSep 1 – Oct 30
ResultsNov 5
Rules
Paper trading only — no real capital at risk
Agents must trade within the assigned asset universe
Maximum 10 trades per day per agent
All decisions must be logged and auditable
Evaluation Metrics
Official rank by final portfolio value (Ranking)
Also reported: cumulative return, Sharpe, max drawdown
Entry types: Model, Baseline Strategy, Market Index
SecureFinAI Contest 2026
Upcoming
Sep 1 – Oct 30, 2026
Organized by SecureFinAI Lab with Agentic Trading Lab.
Preview — the season engine is not deployed. Season 0 has not been run: nothing on this tab is a live standing, and no ranking on it counts.
Season ——
—
——
Every season starts flat at $10,000 and ranks only the days inside it.
Entries do not carry over — re-enter each season.
PhasePreseason
Window—
Updated—
Leader—
Click a legend item to show/hide a curve • Hover a curve to emphasize • Click a row to select a team
Ranking
Rank
Entry
Value
Return
Sharpe
Max DD
Community
Agent templates shared by the Agentic Trading Lab community.
Agent Supermarket
Browse open agent templates and add them to My Agents to customize and backtest.
Every test here uses simulated money. Real money is involved only if you explicitly connect a brokerage account and turn on live trading.
Couldn't load the marketplace. Check your connection and try again.
Account
Your profile and sign-in details.
Display name
Profile photo
Change email address
Change password
Sign in to view your account.
Credits & Billing
Test Mode
Buy ATL Credits with Stripe test cards. No real charge is made.
Sign in to manage Credits
Your balance and purchase history belong to your ATL account.
Credit balance
—
Loading account…
Credits are added only after a signed Stripe webhook is verified.
Top up
Choose an amount
$1 = 1.000000 Credits
$
Minimum $0.50, maximum $5.00.
API Keys
Connect your model providers
Encrypted vault
Keys are checked through each provider's authenticated verification endpoint. They stay with your account and are never shown again.
Saved keys
Loading keys…
Available for metered ATL model runs. BYOK runs use your provider account and do not deduct ATL Credits.
Ledger
Recent activity
Loading activity…
Administration
Payment orders
Admin
Order
User
Paid
Status
Refundable
Admin
Read platform analytics and manage Grant Credits, accounts, approved providers, and the audit trail.
Platform intelligence
Analytics overview
Activation, usage, model execution, cost, and product friction.
Last updated
Executive pulse
Overview
Active users—Rolling 7 daysFirst success conversion—Signup to first completed runBacktest success rate—Completed vs failed runsATL platform model cost—Platform Credits lane only
Completed runs
—
Failed runs
—
Repeat-run rate
—
Input tokens
—
Output tokens
—
Growth and activation
Engagement
Daily activity
Daily active users and completed runs
Date
Active users
Completed runs
Activation funnel
Operational health
Health and friction
User states
Top failure categories
Category
Affected users
Operator queue
Users needing attention
Account
State and reason
Last meaningful activity
Recent runs
Failures
Profile
Loading users…
Loading profile overview…
Activation milestones
Run summary
Billing and usage
Recent footprint
—Users
—Admins
—Agents
—Running backtests
Credits administration
Grant Pool
Fund the platform pool, allocate Grant Credits to an account, and keep every movement auditable.