What should I do today
Checking your coins…
How to read this page. Each card is a ready plan for one coin: what price to enter at, where to get out if it goes wrong, and where to take the profit. The amounts are in money, worked out for your size. Nothing here is advice — the platform does not know the future and does not pretend to.
Loading…
My pairs
Loading…
All markets
Loading…
Recent events
Loading…
What this is. The price of one coin over time. Each bar is an hour, a day, or whatever you pick. It is here so you can check with your own eyes what the card in “Today” tells you in words.
—
Price—
1 hour—
24 hours—
7 days—
RSI 14—
Candles—
—
What this is. Here the platform learns from one coin's past and is then tested on a period it has never seen. If it comes out ahead on that test after fees, the signals for that coin become more trustworthy. If not, the card in “Today” says so honestly.
How to read the result. The only number that matters is edge — accuracy minus 50%. A realistic range for a model like this is 0–3%. Anything above 10% almost certainly means future-information leakage or overfitting, not a discovered pattern. The validation is walk-forward: the model trains on the past and is tested on the future, with a buffer zone cut out between the two.
No model loaded.
Train many pairs at once
What criterion they're selected by. By 24-hour turnover, not by market cap. Turnover tells you two things at once: that the price is being moved by real trades, and that you can get in and out without moving the price yourself. Market cap is a promise, turnover is a fact — a token can have a billion in market cap and twenty thousand in turnover.
The second criterion is history: under 1000 candles, walk-forward validation has nothing to split.
The second criterion is history: under 1000 candles, walk-forward validation has nothing to split.
What this is. “What would have happened if I had followed this method these past months.” Useful for throwing out a bad method. It is not a promise: the past does not repeat on demand.
Out-of-sample vs in-sample. With out-of-sample, the model is retrained for each period using only the data that precedes it — exactly as it would work in real time. With in-sample, the model has already seen the entire period during training, so its numbers look excellent and mean nothing. The difference between the two tells you how much of the "edge" is real and how much is memorization.
No result yet.
Every signal closes itself. When the price touches the target or the stop, the platform records it — losers included. The statistics in the "Results" tab are calculated from all of them, with no cherry-picking. That's the difference between measuring and claiming.
Loading…
Loading…
Strict by default. The thresholds below let through few signals, but only ones with a measured basis. Loosen them and you'll get more — and most of it will be noise.
General
Automatic signals
Timeframe
Minimum confidence
(0.5–0.99)
Also include young markets
Minimum candles per market
Only models with a measured edge
Minimum model edge
%
Non-model confirmations
Trend direction (EMA24/96)
Volume above average
The target is achievable at the current ATR
How many of them are required
of 3
Levels and frequency
Stop
%
Target
%
Cooldown after a signal for the same symbol
hours
Only on direction change
Maximum per day
Quiet hours (UTC)
from
to
Language models don't predict prices. There's no model that can read a chart and know where it's heading — no matter what the marketing promises. So here their role is narrowed to two things: putting into words a signal already made by the statistics, and, optionally, stopping a signal with obviously bad context.
The veto is measured, not trusted. When it's on, some of the signals go through without it and are counted separately. In the "Results" tab the two groups sit side by side. If the filtered ones don't do better than the control ones, the filter isn't doing its job and gets switched off.
Providers
…
How it's configured
The keys go in .env and apply to the whole installation: ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, PERPLEXITY_API_KEY. The explanation is turned on with AI_EXPLAIN, the veto with AI_VETO, and the try order with AI_ORDER. After a change, the platform restarts.
Telegram
…
Webhook / Discord
The signal arrives as JSON. If the address is a Discord one, ready-made text for the channel is added too.
Connected channels
…
New tokens are the riskiest asset category. Thin liquidity, wide spreads and frequent pump-and-dump schemes. The scanner shows what's moving — not what to buy. The trading engine refuses symbols younger than 7 days by default.
Not scanned yet.
How it works. Once you add an on-chain token here, it becomes a full-fledged symbol in the platform — it appears in the chart, the model and the backtest right alongside BTCUSDT. Candles come from GeckoTerminal, prices from DexScreener, and both go into the same cache.
Loading…
The source is unofficial.
api.virtuals.io is their site's backend, not a documented public API — there's no SLA, and the schema can change without warning. That's why dollar prices come from DexScreener, and the platform checks whether the expected fields still exist.
Click "Load".
Trenchor has no public API. There's no documentation, SDK or developer section — just a web app. So the platform reads the PoolFactory contract logs on Base directly and confirms candidates via DexScreener. The method is approximate: it works without an ABI, but it can miss a launch or show an address that isn't a token.
Tokens minutes old are the riskiest asset that exists. This list is for watching, not for entering.
Tokens minutes old are the riskiest asset that exists. This list is for watching, not for entering.
Not scanned yet.
What's available and what isn't. Different sources come with very different guarantees. Here you can see what's currently working and what can be relied on.
Checking…
What this is. Your practice trades and how they turned out. Nothing here reaches an exchange — this is where you check whether your method works before you risk real money.
Positions
No open positions.
Risk limits
Orders
No orders.