Last updated 29 July 2026
Support
01Raising a request
You will be asked to sign in with Google first. That is not a gate on help — it is so a reply reaches you and so your request is attached to your account rather than to an address anyone could type.
What makes a request answerable in one round rather than four:
- What you ran — the exact command, and which service if you know it.
- What happened — the error text, copied rather than described. Log lines are welcome; they contain no credentials, by design.
- What you expected instead. This is the one people skip, and it is often the whole answer: a surprising behaviour and a defect look identical in a log and completely different once the intent is stated.
- Your installation name and roughly when it started.
Never send us a broker key, a password or an API secret. We do not need one to answer anything, and we cannot use one — your credentials live encrypted on your own machine and never reach us. If you have already pasted one anywhere, rotate it with your broker rather than telling us.
Urgent and trading-related? Trip the kill switch first and ask afterwards. Stopping is immediate, needs no confirmation, and is always reversible; waiting for a reply while something is running is not.
02Writing your first strategy
A strategy is a Python class you write, in a folder on your own machine that AlgoSutra mounts read-only. We supply the engine, the risk gates and the data plumbing; the ideas are yours, and we never see them.
The shortest path from nothing to something running:
- Scaffold it —
algosutra-newwrites a working strategy directory, including an instrument ladder, so the first thing you edit is already loadable. - Test it without a market — the toolkit's
Harnessruns a strategy in plain Python with no broker, no data feed and no other service. Three lines, and it needs no test framework. - Backtest it — against your own stored history.
- Run it on paper — live data, simulated execution, realistic slippage and charges, for as long as you like.
Two things that surprise people early, both deliberate. Your strategy receives ticks, not bars — the toolkit builds bars for you, because how you want them bucketed is a decision only you can make. And between emitting an order and hearing its fill, your own position ledger still reads flat; a strategy keyed only on "am I flat" will re-emit on every tick until the fill lands.
03Choosing features
The platform ships a catalog of computed features — momentum, volatility, microstructure, option-surface and more — plus a parametric layer that builds any classical rolling feature at any window you name.
The distinction that matters when you are choosing: not every feature can be computed live. Some need a session to be over before they mean anything, and some need data that only arrives in a backfill. The toolkit refuses those at construction rather than returning nothing forever, and tells you which they are — so if a feature is rejected when you build your indicator set, that is the answer, not a bug.
If you want a feature we do not ship, write it. A folder of your own feature code is mounted the same way strategies are, one file per feature. Keep the maths inside the class: only first-party source is hashed into the run manifest, so a helper defined outside it is outside the record of what produced a result.
04When something looks wrong
Before anything else, the console tells you what the platform thinks of itself. Three things are worth checking in order:
- Is the data fresh? A stale feed shows a visible disconnected state and halts dependent strategies rather than acting on prices that have moved. If strategies have stopped, this is the first thing to rule in or out.
- Is the broker connected? Broker sessions expire and are re-established at start-up; a login that lapsed overnight looks like a platform fault and is not one.
- What does the audit trail say? Every order, fill, configuration change and control action is recorded and attributable. It answers "did it do the thing" far faster than reasoning about whether it should have.
"It stopped trading" is usually the platform working: a risk gate refused an order, or freshness halted a strategy. The logs name which gate and why.
05Licences and renewals
Your licence is a signed file your machine verifies on its own, against a key it already holds. If this website is down your platform keeps running and keeps trading — licensing is never in the path of an order.
Because it is signed, it cannot be edited in place. Renewing issues a new one, and you replace it on your machine; your account page is where you fetch it.
A lapsed licence stops new positions being opened. It never stops you closing one, and it never blocks the kill switch — a commercial dispute must not be able to trap you in a trade.