The whole journey in order. Each step is deliberately gated by the one before it — that's not friction, that's the product.
Go to /grade — it's public, free, and doesn't ask you to sign up.
e.g. “Buy SOL and ETH when RSI(14) drops below 30 on the 1-hour, take profit at 2%, stop loss at 3%.” Be specific about timeframe, entry, and exits — whatever you leave out, the engine will assume and tell you so.
In about a minute you get a letter grade, the five gate checks, and the honest numbers: net expectancy after fees, R:R, win rate, drawdown — measured out-of-sample on 6 years of data.
Every grade gets a permanent public URL. Share it, or come back after reading the next tutorial.
Five hard checks (enough out-of-sample trades, positive net expectancy, drawdown in bounds, robustness, execution realism) — pass all five or the agent can't trade real money on Tessen. There is no negotiating with the gate.
The average trade's profit after fees, in basis points. This is the number that decides everything; if it's negative, more trades just means faster losses.
76% win rate with 1:0.3 R:R loses money. 40% win rate with 1:2.5 prints. The report always shows both — read them as a pair.
All headline numbers come from the 40% of history the strategy never saw during design. In-sample results are how backtests lie; we don't show them as the headline.
Each report includes a deterministic diagnosis of what dragged the grade down — the specific check that failed and by how much, not a vague score.
Tessen is in closed beta — join the waitlist on the homepage, or use your invite code at sign-up.
In the Studio, describe the strategy conversationally or configure signals directly: indicator rules (RSI, MACD, Bollinger, stochastic…), a timeframe, exits (take-profit / stop / trailing / time), optional regime filters and higher-timeframe confirmation.
Every agent must be graded before it can do anything else. The grade panel shows the same honest report as the public grader.
Change one thing at a time and re-grade. If you find yourself hunting the one parameter combination that turns the grade green — stop. That's overfitting, and the out-of-sample split will catch it, but your future self pays for it either way.
From My Agents, deploy the agent as a paper deployment — it runs on live market prices with simulated fills and a simulated balance.
A backtest compresses years; paper trading is the slow-motion check that execution, timing, and your own nerves behave. Two weeks is a reasonable minimum.
The live dashboard tracks equity, open positions, and drift — whether live behavior is diverging from what the backtest promised.
The grade predicted an expectancy. Paper results should live in the same neighborhood. If they don't, believe the paper results.
On your exchange, create a new API key with reading and trading enabled and withdrawals disabled. Never reuse a key that can withdraw.
Keys are encrypted at rest and used only to place the trades your deployments decide on.
Tessen checks the key against the exchange's actual permission response — a key with withdrawal rights is rejected outright, for your protection.
Per-exchange walkthroughs (Binance, Bybit, OKX, Bitget, Coinbase, Hyperliquid) are on the exchange guides page.
Only gate-passing agents can be deployed with real money. This is the point of everything above.
In confirm mode the agent does everything except pull the trigger: on a signal you get a Telegram message with the exact entry, size, stop and target, and Approve / Reject buttons. Unanswered proposals expire safely.
Positions are sized from your risk settings with hard caps, and every deployment carries automatic halts: daily-loss stop and max-drawdown stop.
After confirm mode has shown you enough good proposals, switch to live mode — same engine, no approval step. You can drop back to confirm at any time.
Open @tessenaibot on Telegram and link it from your Studio settings — one code, one tap.
Trade proposals and fills, Lumen market-signal alerts (with per-signal and per-asset opt-in), and pass expiry reminders.
Lumen alerts fire when a signal enters an extreme state — not every five minutes while it stays there. Silence means nothing changed.
Studio → Developer. The raw key (tsn_live_…) is shown once at creation; we store only a hash.
POST /v1/api/grade/spec with an AgentSpec JSON — same engine, same honesty, returns the grade and a verify link. Or POST your own trade history to /v1/api/grade/record for a “Checked” grade of real past trades.
GET /v1/api/lumen/snapshot for the current cross-venue signal state.
Free tier: 100 calls / 30 days. Developer tier ($29, prepaid, no auto-renew): 5,000.
Full endpoint reference with schemas: /docs/api.