TTESSEN/ docs

User Guide

The complete walkthrough — from your first prompt to a live, monitored agent. Fifteen minutes to read; it covers everything the platform can do today and is honest about what it can't.

1. Getting access

Tessen is in closed beta. You need an invite code to activate your account: join the waitlist and codes go out in batches as seats open. Once you have one, sign up at /sign-up (email/password), enter the code when prompted, and you land in the Studio.

2. The Studio — describe an agent in plain English

Type your strategy idea the way you'd explain it to a person: "Buy SOL when RSI(14) drops below 30 on the 1-hour, take profit at 3%, stop at 1.5%." Tessen drafts a complete AgentSpec — a precise specification where every parameter is explicit. Two things to check before grading:

  • Assumptions — anything you didn't specify, the AI filled in and tells you so. Read them; change what's wrong.
  • Questions — genuine ambiguities it wants you to resolve rather than guess.

The building blocks available today: indicators rsi, ema, roc, cmf, adx, atr, bb_pband, macd_hist, stoch; exits by take-profit %, stop-loss %, time, ATR trailing stop, breakeven arming, and one-time partial profit lock; regime filters (ADX floor, ATR-volatility terciles); higher-timeframe confirmation; and multi-sleeve agents that run different logic per timeframe. If you ask for something outside this set, the AI says so instead of silently approximating.

3. Reading your Grade

Hit Grade and the engine backtests your spec on real historical data — out-of-sample, net of fees, with realistic fills. The letter (A–F) is decided by a hard gate of five checks. Fail any one and the grade is F, no matter how pretty the backtest looks:

  1. Positive out-of-sample expectancy — it makes money on data it never trained on.
  2. Clears the cost hurdle — the gross edge is at least 1.5× total trading costs. An edge thinner than fees isn't an edge.
  3. Robust across assets — at least 60% of the coins it trades are individually profitable. One lucky coin doesn't count.
  4. Survivable drawdown — at responsible position sizing, the median worst drawdown stays under 35%.
  5. Not overfit — out-of-sample performance holds at least half of what training data promised.

The headline number is net expectancy in basis points per trade— never win rate. A 78% win rate with a bad risk/reward loses money; we've proven that on our own strategies. Win rate is shown as context only. Below the letter you get per-asset results, the sizing math, and "Why this grade?"— a plain-English post-mortem of what specifically failed and what to try changing. Expect D's and F's: most strategies fail, and finding that out for free is the product.

4. Iterating and saving

Change one thing at a time — "widen the stop to 3%" — and re-grade. The AI amends only what you asked and tells you what it kept. Save graded agents to My Agents; each card keeps its full Grade report, and you can reopen any agent and continue iterating. Add journal notes to a card to record why you made a change — future-you will thank you.

5. Paper trading — free, no keys, zero risk

Any saved agent can be paper-deployedfrom its card. It then evaluates live market data every minute using the exact same entry/exit logic as the backtest — real current candles, simulated fills, no exchange account needed, no real orders ever. The card shows a live tally of closed paper trades. One honest disclosure: the backtest derives its volatility-regime thresholds from a train/test split that doesn't exist live, so paper mode approximates that one filter with a rolling window. Everything else is bit-identical logic.

6. Connecting an exchange

Under Connect Exchange, add a Binance Futures API key. Non-negotiables, enforced in code:

  • Keys with withdrawal permission are rejected at connect time. Create a trade-only key.
  • Tessen never holds your funds. Your money stays on your exchange account; we can only place and close trades.
  • Keys are encrypted at rest (AES-256-GCM) and never shown again after connect — only a masked preview.
  • Recommended: IP-restrict the key on Binance's side too. Belt and suspenders.

7. Going live

Live deployment is deliberately gated. An agent must pass all five Grade gates, and real-money mode additionally needs a connected futures key and an active Pro pass. The path we recommend (and follow ourselves): grade → paper → dry-run live → live. Dry-run runs the full live engine — real signals, real sizing math, simulated $1,000 equity — without touching your account, and needs no key or pass.

When real money is on, every trade gets:

  • Risk-based sizing — position size is computed from your risk-per-trade % and the stop distance, not a fixed bet.
  • Leverage from the stop, hard-capped at 10×, with liquidation kept at least 2.5 stop-distances away. Isolated margin.
  • Exchange-resting stop-loss and take-profit orders — protection lives on the exchange, not in our scheduler. If placing the stop fails, the position is closed immediately rather than left naked.
  • Automatic halts — a daily-loss limit and a max-drawdown limit that flatten everything and stop trading. Halts never auto-resume; you decide.
  • A kill switch — one click flattens all positions and stops the agent, any time.
  • Telegram alerts for entries, exits, halts, and drift (link your account via @tessenaibot).

8. The Dashboard and grade drift

Dashboardshows each deployment's equity, today's P&L, drawdown, open positions with live unrealized P&L (click a row for a candle chart with your entry, stop, and target drawn on), and closed-trade stats — refreshed every 30 seconds. The most important chip is drift: Tessen continuously compares live results against what the Grade promised. Under 20 closed trades it honestly says "warming up" instead of pretending to know. If live expectancy falls materially below the promise, the deployment is flagged drifting and you get one clear Telegram alert — the platform that graded your agent is the same one that tells you when the grade stops being true.

9. Billing

Building, grading, and paper trading are free. A Pro pass unlocks real-money live trading: $49/30 days, $129/90 days, $449/year — prepaid, no auto-renew, no card on file. Passes stack (buying early extends your expiry). Pay in USDT on TON, other coins, or by card. See Pricing.

10. NEXA — copy agents, black-box

Only gate-passing agents can be listed on NEXA. Copying runs the agent's signals on your own exchange keyswithout revealing its logic — the owner's strategy stays private, your funds stay yours. Cost is a performance fee on real profit only: a losing period owes nothing. Every listing links its verified Grade — check the sample size and drift status before copying, exactly as you would your own agent.

11. Lumen and Aurix

Lumen is cross-venue market intelligence — positioning, funding, open interest across 10 exchanges — with every signal labeled either scan-validated (survived weekly robustness re-testing) or information-only. Aurix is trader memory: post-mortems on every grade, journals on every agent, and behavior analytics across your history — the layer that turns losses into lessons instead of repeats.

12. Honest expectations

  • Most strategies grade D or F. That is the system working, not failing.
  • A passing grade is a measured historical edge, not a profit promise. Markets change; that's what drift detection is for.
  • Never allocate money you can't afford to lose, even to an A. Position sizing and halts limit damage; nothing eliminates risk.
  • If anything on the platform ever seems to promise profits, report it — that's a bug in our copy, not a feature.

Questions not covered here: FAQ · How the Grade works · Security · Developer API