Hyperhelm is a governed decision cockpit. Three engines produce one gated trade verdict, and you execute it non-custodially — perps on Hyperliquid, spot on CoW. You sign every order; Hyperhelm never holds your keys or your funds.
A trading agent (or you) shouldn't act on a setup without knowing how much risk is allowed. Hyperhelm answers that: it reads market structure, matches a playbook, and the Risk Governor sizes the trade. The verdict — direction, size, leverage, entry/stop/target — is what you deploy. The governor sizes the order; that is the whole point.
Click Connect wallet (top-right). A short explainer appears, then your wallet asks to connect to app.hyperhelm.xyz — that's expected and standard: it authorizes the cockpit to read your address and request signatures. It never moves or holds funds. Connecting also lets the cockpit show your open Hyperliquid positions and resting CoW spot orders in the Deploy panel.
Every verdict traces back to three independent engines. Each is a standalone product you can open from the left rail.
Market Structure Engine — breakouts, breakdowns, ranges, cycle context.
Trading Playbook Engine — which playbook fires, its conditions and side.
Risk Governor (Risk Permissions) — the pre-trade verdict, max size, leverage cap, binding cap.
The hero shows the live verdict for the selected asset (BTC / ETH):
riskstate_v2, the portfolio-aware gate.Several non-conflicting playbooks can fire on one asset (say, a defensive trim and a structural accumulation). The Strategist card lists every fire with its side and the Navigator's ALIGN / CONFLICT vote, and marks the one chosen by priority. The chips above the verdict let you select any of them:
Cooldown: once a setup fires and alerts, the engine won't re-signal it until its cooldown window resets. A cooldown-locked setup shows its unlock countdown and the Deploy button is held — it's the engine saying "you already had this alert". GATE: OFF (§9) can override it, logged as discretionary.
In governed mode the size slider goes down only, capped at the governor's max size. In override mode (§9) the governor cap becomes a reference and you can size up to the hard safety ceiling.
The chart draws the market with toggleable layers: Verdict (the selected setup's entry/stop/target), Structure (the Navigator's levels — breakout/breakdown triggers, range, major support), and Signals. Switch 1D / 4h for structural context vs near-term break watch. Nothing on it is fabricated — every line comes from an engine.
The Deploy button places your sized intent — entry as a GTC limit, with TP + SL as native reduce-only triggers (OCO). First deploy of a session asks for one wallet signature to approve the agent; after that, orders sign silently in-tab.
Open positions show live uPnL / ROE / leverage right in the Deploy panel, with close-at-market one click away (mainnet asks to confirm). HL is one-way: deploying on an asset you already hold nets into that position.
The Trade spot on CoW button opens the embedded CoW swap with a chain selector (Ethereum / Arbitrum / Base / Gnosis). A banner shows the setup's size and entry; you enter the amount and, for an entry price, set the limit yourself (CoW does its own price discovery). Your pending CoW limit orders show back in the cockpit's Deploy panel, next to your perp positions.
The top-bar toggle controls how binding the gate is. It is venue-agnostic: a BLOCK holds both the Hyperliquid CTA and the CoW CTA — the gate is about adding exposure, not about a venue.
Hyperhelm charges only on the execution edge, usage-based, shown transparently before you sign — no subscription, no wall:
The track record splits by venue: SPOT · CoW and PERP · HL, each with Open · Latent · Closed (W/L) · Realized · WR. Latent is the marked-to-market return of open exposure (spot accumulation's edge lives here); Realized is closed round-trips. A live disposition line explains the "fired but not opened" gap: how many setups are blocked by the gate, not chosen (conflict), structure-blocked, cooldown-locked, or ready to open.
The benchmark compares the governed book against the ungoverned counterfactual — what the same fires would have returned with no governor. Five tiles: Helm · governed, Ungoverned, Governor Δ (governed − ungoverned), Drawdown saved, and Losses avoided (with the count of blocked fires). The governor's value-add is measured, not asserted — and overrides never masquerade as governed results.
?net=testnet to the cockpit URL to drive the full flow on Hyperliquid testnet without real funds.