Frequently asked questions

Below are the questions we get asked most, answered plainly. Coil is one brain sold three ways: Coil Scanner is the scored board you read yourself, Coil Live is the same board served to your own AI agent over MCP, and the Coil Engine is the whole system as software you download and run. Several answers come down to the same idea: you own the keys, the capital, and the risk. Nothing here is investment advice or a recommendation to buy, sell, or hold any security. If you want the longer walkthrough first, the how-it-works page covers the mechanics end to end.

The basics

Do I need my own AI agent?

For the Engine and for Coil Live, yes. Coil is the brain and the rule engine; your agent is the hands. For Coil Scanner, no — you read the board in a browser and place any trades yourself. It is built and tested for Claude Code running as a scheduled agent against a brokerage connector (built for Robinhood's agentic accounts over MCP). If you don't have an agent and don't want one, the Engine is not for you — Coil Scanner is the same scores in a browser tab, and you place any trades yourself.

Does Coil place trades by itself?

No. Coil contains no order-placement code and never holds your broker credentials. The engine decides and writes the order; your own AI agent reads it and places it at your brokerage. Two consequences: Coil cannot trade without an agent you connect, and it ships disarmed — going live is a separate gated step (allowlist your account, re-pin the integrity fingerprint, pass the self-test, type a total-loss acknowledgment).

What is Coil?

Coil is a rules-based, long-only stock scanner and trading engine that finds market leaders at real pullbacks — and can execute those same rules through your own brokerage account, run by your own AI agent. It is sold three ways: Coil Scanner (the board, for you), Coil Live (the engine-grade feed, for your agent, over MCP) and the Coil Engine, which you download and run yourself on your own Mac or Windows machine, inside your own AI agent, with your own capital. The Engine is three things in one download: a scanner that scores every name in the Equities book (S&P 500, Nasdaq-100, and Macro — ~560 names); a dashboard that shows you the whole board; and a long-only engine that trades those scores by rule. It is not a fund, not a managed account, not a signal service, and not investment advice. You own the keys, the capital, and the risk.

Is Coil legit and safe to use?

Fair question — the AI-trading category is full of scams and you should assume the worst until shown otherwise. The short, checkable version: Coil never takes deposits, never holds your money or your broker credentials (the engine runs on your machine; your capital stays at your brokerage), sells through Gumroad as merchant of record, guarantees nothing, and publishes its scores hash-committed before outcomes are known — plus a forward-return audit of its own signals even when it reads negative, which right now it does (about 0.2pp behind the sector control). The full regulators'-checklist walkthrough, with a live link for every claim, is at coil.trade/trust.

What does Coil trade?

Market-wide leaders. Coil scores every name across the S&P 500, the Nasdaq-100, and a Macro book (bonds, income, gold and metals, commodities), and buys whatever is actually leading — at a real entry. It is long-only. For the highest-conviction leaders it can accelerate the position through a mapped leveraged ETF (for example NVDA via NVDL, or the Nasdaq-100 via TQQQ) at reduced size, so leverage is an accelerator on the best names rather than the default vehicle. When semis lead, Coil holds semis; when leadership rotates, so does Coil. The board Coil publishes also carries the Crypto book (BTC/ETH) as a read — the engine itself does not trade crypto.

Does Coil trade crypto?

The engine, no — the $79 download is a long-only equities engine and does not trade crypto. The board, yes as a read: the Coil Scanner board carries a Crypto book that scores the 81 Robinhood-tradable pairs around the clock, and since July 15, 2026 it has published a long-only BTC and ETH trend signal — two 50% sleeves, each in-asset only while its price holds above its 50-day average and BTC holds above its 200-day, in cash otherwise, evaluated on the UTC daily close. Signals are tracked live since July 15, 2026 — backtest before that — and the Crypto book is not traded with real money by Coil. The full numbers, losses included.

Do the crypto signals beat buy-and-hold?

No — and we lead with that. Over 2016–2026 the cold backtest compounded 59.6% CAGR versus 74.9% for a 50/50 BTC/ETH hold. What the rule buys is smaller crashes: 2022 finished 0.0% versus −66.0% for holding, with a max drawdown of −64.3% versus −87.4% — at the cost of whipsaw years (2024: −22.6% versus +74.7%). It is a risk posture, not a moonshot; these are research backtests under modeled conditions and do not predict future results. The record, ugly parts first.

Is Coil investment advice?

No. Coil is software, not advice and not a recommendation to buy, sell, or hold any security. It does not know your finances, goals, or tax situation, and it makes no suitability judgments. It executes pre-written rules against live prices. Everything it does happens in your own account, under your control, and the outcomes are yours. If you want to understand how the rules differ from a hand-tuned script, see self-tuning vs static trading bots.

Can Coil short the market?

No — Coil is long-only by design. There is no short book. When the market turns down, "down" means raising cash and rotating defensively into the Macro book (bonds, income, gold), not shorting. This is a deliberate choice: shorting inverts risk and adds a failure mode we would rather not carry, and cash is a perfectly good position.

What happens when nothing qualifies?

Cash is a position. Coil does not force a trade to look busy — if few leaders are at a real entry, or the market posture is defensive, it raises cash and waits. Any uncommitted cash earns whatever your broker's variable sweep program pays (for example, Robinhood Gold quoted about 3.35% APY in early 2026 — that is the broker's variable yield, not paid by Coil and not risk-free). Standing down is just one of the moves the rules can make.

Cost and billing

What is Coil Live?

Coil Live is the tier built for software rather than for a person. It serves your own AI agent the engine-grade board the live record trades from — per-name structural stops, targets, conviction and lanes — plus the served trading discipline: the rules the engine itself follows, delivered as a rulebook your agent reads rather than logic you have to write. It also serves a ready-made candidate list, so your agent is handed the entry pool instead of deriving it from raw fields and getting it wrong. It costs $29/month or $249/year, connects over MCP or plain HTTPS with a license key as a header, and refreshes every 5-9 minutes in market hours. Your agent runs on your machine and places every order at your own broker; Coil never sees your account, your keys, or your positions.

How do I connect my agent to the MCP feed?

One header. Add https://coil.trade/mcp to your agent as an MCP server with the header X-License-Key: your-key (the key arrives with your Gumroad receipt), or call GET https://coil.trade/api/live/feed over plain HTTPS with the same header. Then tell your agent to call get_live_setup and follow it. The served rulebook’s first rule is dry-run: print intended orders, place nothing, until you deliberately arm it. Coil cannot enforce that on your machine and does not pretend to; the arm switch is yours. The Coil Live page shows the exact command.

Should I buy Coil Scanner, Coil Live, or the Coil Engine?

Buy Coil Scanner if you place your own trades and want to read the board yourself — it is the full scored market in a browser tab for $12/month. Buy Coil Live if you already run an AI agent and want it to act on the board without you writing the strategy — it serves the feed and the discipline to that agent for $29/month. Buy the Coil Engine ($79 once) if you would rather own the whole system as source on your own machine and never pay again. The scanner brain is the same in all three; what differs is who reads it — you, your agent, or your own copy of the engine.

Can I try Coil before paying?

Yes, three ways, none of them a trial you have to cancel. The Scanner page shows a free preview of the board (top names per book, one day delayed) and the demo is the real dashboard with frozen data. For agents, the get_live_setup tool on our MCP server answers with no key, and the free delayed board is at /api/board/free. The rules themselves are public on how it works. There is no time-limited trial; month one of a subscription can be cancelled the same day from your Gumroad library.

How much does Coil cost?

Three products, one brain. Coil Scanner ($12/month or $99/year) is the AI-powered scanner for traders who place their own orders — the full board in your browser at coil.trade/scanner, re-scored every ~5 minutes through the market day. Coil Live ($29/month or $249/year) serves the engine-grade feed — structural stops, targets, conviction, lanes — plus the served trading discipline to your own AI agent over MCP at coil.trade/live-feed; your agent places the orders at your broker and we never see your account. The Coil Engine ($79, one-time) is the whole system as source on your machine — yours to keep and run forever, no recurring charge.

Is Coil a subscription? How do updates work?

The engine is not a subscription — it is a one-time $79 purchase you keep forever, offline, with no recurring charge. New versions are announced on coil.trade and installed manually when you choose to; nothing is auto-pushed. Coil Scanner ($12/mo or $99/yr) and Coil Live ($29/mo or $249/yr) are subscriptions — hosted publications re-scored through every market day, so hosting and publishing are ongoing work. Both renew until you cancel from your Gumroad library, access runs to the end of the paid period, and partial periods are not prorated; billing errors are always corrected.

What is Coil Scanner, and how is it different from the engine?

Coil Scanner is the scanner’s live output, hosted: every name in the Equities book (S&P 500, Nasdaq-100, and Macro) — plus the Crypto book (81 Robinhood-tradable pairs, around the clock) — re-scored about every 5 minutes through the US market day, with regime verdicts and buy-list lanes, readable in your browser — $12/month, cancel anytime. The engine is the whole system you download once ($79) and run inside your own AI agent against your own brokerage; it includes this same scanner locally. Scanner = the read, for humans. Live = the same feed, served to your agent. Engine = the system, on your machine. If you buy the Engine you don’t need the Scanner; a Live key opens the Scanner too. The Scanner is an impersonal research publication — same board for every subscriber, no stop/target prices, not individualized advice.

Can I get a refund?

All sales are final. Coil is an instant digital download you keep the moment you buy it, so once it is delivered there are no change-of-mind refunds. The one thing we always fix is a genuine billing error — a duplicate charge, or being charged the wrong amount — which is a billing correction, not a refund. Checkout, receipts, and license keys are handled by Gumroad as merchant of record; email support@coil.trade for a billing correction and we will sort it out. Full details are in the refund policy.

Running it

Do I need Claude to run Coil?

Coil is built and tested for Claude running as a scheduled agent (Claude Code), and that is the supported, recommended setup. The engine itself is plain Python, so a capable agent that can run code on a schedule and call your broker connector could in principle operate it — but Claude is what we test against, and we make no guarantees for other agents. For more on what an agent can and cannot do here, see can Claude trade stocks?

What do I need to run Coil?

Four things: an AI agent (Coil is built for Claude running as a scheduled agent), a brokerage connector (built for Robinhood, works with any equivalent broker MCP), a Mac or Windows machine with Python 3.9 or newer, and a free Alpaca market-data key for price history (the scan only — trading and live intraday data run on your broker's own connection). Robinhood officially supports agentic trading with dedicated agentic accounts connected over MCP — see Robinhood's agentic-trading page. Setup is agent-led: point your agent at the folder and it runs the interactive setup, introduces itself, scans the market, and opens the dashboard with live scores within minutes — then watches in dry-run. Going live is a separate, deliberately gated flow. The Claude + Robinhood agentic trading guide walks through it.

What happens if my computer is asleep or offline?

Nothing runs, and that is the trade-off of software you own. Coil acts only while a scheduled agent session is running on your machine — there is no Coil server acting on your behalf, which is exactly why we never hold your keys or your capital. It also means Coil places no resting stop orders at your broker: stops and targets are evaluated each cycle by the engine on your machine. So if the machine is asleep, powered off, or without internet during market hours, no cycle runs and nothing is managing your open positions until it wakes. What you hold stays held — stopping never sells what you hold — but it is unmanaged in the meantime, and a position can move against you while nothing is watching. If you want coverage without keeping a laptop awake, run it on a machine that stays on.

Whose money and credentials are these?

Yours, entirely. Coil trades your own capital in your own brokerage account. Your broker credentials and API keys stay on your machine and never leave it — there is no Coil server holding your keys, your money, or your positions. Coil has no ability to move funds out of your account, and it can only touch the single account you explicitly allowlist. There is no deposit, no vault, and no withdraw button anywhere in Coil — nothing to fund, nowhere to send money, and no code that could. Read the full custody architecture, with the code receipts.

Risk and results

What stops someone extracting the strategies — or the AI leaking how it all works?

With Coil there is nothing to extract, because you already own it. The strategies ship inside the packet you buy, as readable code on your machine — that is the product, not a leak. Hosted platforms have to defend a server that holds their strategy, their customers' credentials, and sometimes their customers' money; a clever prompt or a breach there has something to find. Coil's hosted surfaces serve published board data only: they hold no credentials, no customer funds, and no code that can move money. Your broker keys live on your machine and never leave it. Our servers cannot leak what they never hold.

Can Coil lose money?

Yes. You can lose money — including, on any leveraged position the engine takes, the total loss of the amount committed to it. No setting, circuit-breaker, or stop changes the fact that markets can move violently against a position, and leveraged ETFs are designed for short holding periods. The honest framing on leveraged-ETF decay is worth reading before you run anything live.

How does Coil manage risk?

Two ways. First, entry discipline: it only buys leaders at real entries (scored READY or SETUP), never chasing extended names pinned at their highs and never buying falling knives, and it sizes leveraged vehicles down. Second, a hard safety spine underneath the strategy: an account whitelist, hard position and exposure caps, a drawdown ladder keyed to your equity high-water mark, a kill switch, and fail-closed behaviour — if anything is ambiguous, it does nothing. These reduce and bound damage; they cannot guarantee an outcome, and stops can still gap straight through a price.

What are the backtested results, and how trustworthy are they?

Coil's scoring is built on a research backtest, not a live track record. On point-in-time index membership (delisted names included, so no survivorship bias), with next-open fills and costs modeled, the leadership-rotation backbone compounded +638% from 2017 through the first half of 2026 versus SPY's +282%, with a shallower worst drawdown (−23% vs −32%) and a positive result in 9 of 10 years (worst −1%, in 2018). The honest rider matters, so we lead with it:

Where is the live record, and what does it leave out?

At coil.trade/results: one real account, our own capital, running these rules, published every settled session win or lose against SPY and QQQ, with a by-month table and a ledger of every closed round trip, named. Two things are excluded and both are printed there: sessions the engine was not running (held flat, not graded) and the handful of operator-placed trades the engine never chose, backed out by amount. The record is evidence that the rules work when followed; it is not a projection of any other account. The same data is free as JSON at /api/perf and /api/trades.

Will my results match the live record?

No, and that gap is not small print. Your copy or your agent reads the same board and the same rules, and still prints different numbers: it polls at different moments and sees a different candidate list; it sizes every position against your equity, not ours, which can correctly mean no trade at all; it fills at your broker, at your moment, across your spread, under your own settled-cash and pattern-day-trader limits; and if you own the Engine and change a rule, you are running your strategy, not ours. What transfers between accounts is the discipline, not the profit and loss.

What does "self-tuning" actually mean?

The validated rule-set is frozen and fingerprinted — Coil is not an open-ended AI inventing new trades on your machine. What adapts is narrow: every close, a local learning pass leans position sizing (only sizing) toward what has been working across your own closed trades, split by book (S&P 500 / Nasdaq-100 / Macro) and by buy-score tier. The tilts are shrinkage-weighted and hard-clamped (roughly 0.55×–1.35× overall), so one trade barely moves the dial. It never touches entries, exits, stops, or caps and cannot invent new structure, and every tilt is written to a readable audit trail. New validated structure arrives only as a deliberate new version you choose to download. The packet ships the research backtest harness and a safety self-test suite so you can re-run the research and verify the spine yourself. If you want the contrast with a frozen script, self-tuning vs static trading bots goes deeper.

If lots of people run Coil, doesn't the edge disappear?

The honest version of this question deserves an honest answer. Most of what Coil does is not a secret signal that crowds out — it is discipline: sizing positions, refusing extended entries, standing down when the regime says cash, taking losses early. Discipline does not stop working because other people also behave well. The selective part — leadership rotation — has been documented in public research for decades and still persists, and Coil's entire user base is a rounding error against index flows. If our flows ever became big enough to matter, it would show up in the published record first — and we publish the record, graded against SPY, losers included. What we will never promise is that everyone wins: trading has zero-sum aspects, and anyone promising otherwise is selling something else.

AI models change every month. How can any track record mean anything?

That worry is legitimate — an agent whose judgment lives in a chat model's mood has no stable thing to grade. Coil is built the other way around: the strategy rules, the scoring, and every safety rail are versioned software, not model vibes. The AI executes a written procedure; the judgment it follows is code with a changelog. Rule changes ship with a public changelog line, the safety rails are not knobs on any profile, and the graded record runs continuously across versions — so if a change degrades results, the record shows it, because the reds get published with the greens. You are not asked to trust a model brand. You are asked to check a ledger.

Does Coil learn from my trading?

Yes — gently, and only your sizing. Every close, a local pass reads your own closed round trips and leans position sizing toward what has been working, split by book (S&P 500 / Nasdaq-100 / Macro) and by buy-score tier. It is size-only: it never touches which names it buys, the entries, the exits, the stops, or the caps. The tilts are shrinkage-weighted and hard-clamped (roughly 0.55×–1.35× overall), so one trade barely moves the dial and it stays neutral until you have a couple of closed round trips behind it. It learns from live round trips only — nothing tunes until you arm the engine and close real trades — it runs entirely on your machine with nothing leaving it, and every tilt is written to a readable audit trail (optimization_log.json). It also surfaces diagnostics it never auto-acts on: which sectors, sleeves, and entry-quality bands have been paying or bleeding.

Independence

Is Coil affiliated with Robinhood or any index or fund?

No. Coil is independent and not affiliated with, endorsed by, or sponsored by Robinhood, Alpaca, or any exchange, index, or fund it references. It is built to work with Robinhood and Alpaca as a user-operated connector and data source; any tickers it trades are named descriptively, and Coil has no relationship with their issuers.

Still deciding whether Coil fits your situation? It helps to compare it honestly against the alternatives in Coil vs trading bots and signal services, and to read the leverage-decay primer before you commit any real capital to a leveraged position.