/Sign in

Agent

In one conversation, write strategies, backtest, analyze, tune, write factors, read documents and ask about data.

Contents

Write a strategy

Describe a trading idea in one sentence, for example:

The Agent writes it as a strategy and backtests it; then, in the same conversation, change it, analyze it, or tune it. The full steps are under Strategies.

The Agent page: template cards and the input box on the left, the result board on the right

Clicking a template card sends it and starts the run; like your own messages, it counts as a run and uses points. To change the wording, write it yourself in the input box, following the card.

Write a factor

Say, for example:

The Agent proposes a factor (a formula or code); once you confirm, it builds the factor, runs the four gates and interprets the report. You need "factors" on the "Includes" line of Settings → Points, see Tiers and points.

  • Change a factor you have: name a factor in your library and say which horizon, bar size or universe to use; once you confirm, a new version is made and the four gates run.
  • Review: say "review <factor name>", or "review this factor" in the conversation where you just made it; the conditions are under Review.
  • Promote: say "promote <slug>"; the conditions, and when a second confirmation is needed, are under Promote.

The details of each step are under Factors.

Read a document

Click "Add a document" under the input box, or drag a file straight into the conversation, to bring in a paper, a report or a web page; the Agent reads it, picks out ideas and writes a research note. Do not paste text over 2,000 characters straight into a message; bring it with "Add a document" too, or paste it on the "Research" page.

"Add a document" under the input box on the conversation page: click it to bring a file or a link, or drop the file into the conversation
  • Formats: PDF with a text layer, plain text, Markdown, HTML, or a public web link; with "pictures" under "Includes" you can also bring PNG, JPEG or WebP pictures. One document per message.
  • What to look for: whatever you write in the message is the reading brief (up to 600 characters, optional). Send the same document again without a brief and you get the last note; with a brief, it is read again to that brief.
  • A strategy from a document: say "write a strategy from “document title”" and the Agent writes it from that document; if you have several and name none, it asks which one.

You need "research" under "Includes" on Settings → Points.

Ask about data

You can also ask about the market without building a strategy, for example:

The answer uses exchange data, and charts go to the "Chart" tab on the board. If you connect your own database, you can ask about it too, see Connect your own database.

Market and data source

Market: the menu under the input box chooses "Crypto perpetuals", "Crypto spot" or "US equities"; crypto data comes from Binance today; the difference is under Market data on the platform. A strategy trades one market.

  • If the conversation has no strategy yet, just switch.
  • If it already has one, you are asked first: "Make the X version" starts making another version for the new market right away, and the original strategy stays as it is; "Keep Y" does not switch.
  • You can also just say it in a message, such as "do it on spot": while the conversation has no strategy and you have not picked a market yourself, the menu switches to it and tells you; at other times you are asked first.

Data source:

exchange bars

Binance's real price history; the default

Synthetic bars (fast)

Generated prices, good for checking that a strategy runs; results are marked "synthetic" and do not reflect real prices

Your data

Bars you uploaded or imported on the "Data" page that passed validation

Run automatically, or approve first

The run mode under the input box has two settings, "Automatic" by default; once you pick one, this browser keeps it, for new conversations too:

  • Automatic: a new strategy is backtested as soon as it is written.
  • Ask before the backtest: a card lists the symbols, bar size and range before the backtest; click "Approve and run" or "Reject".
With "Ask before the backtest", a card appears before the backtest; it starts only when you click "Approve and run"

Even with "Automatic", a change to the current strategy code, the candidate backtests in tuning, restoring an earlier version, and building a factor or a new version of one and running its four gates show a confirmation card first and start only once you approve.

While a run is going

Above each reply is a progress card with the steps and their times, and a progress bar during backtests. Click "open run" above the card for the full run page.

  • Stop: click "Stop" while it runs. It stops after the current step finishes.
  • The Agent has a question: a yellow box "The agent has a question" appears in the card; answer in the input box and the Agent carries on from your answer (as a new run). Sometimes there are options; clicking one sends it as your answer.
  • Connection lost: the page shows "Connection lost, reconnecting…" and reconnects by itself.

Saving a version, opening a paper account and submitting to a contest never happen on their own; you start them. To save a version, click the button on the board or say "save as a version" in the conversation.

The board

Results show on the board on the right. Under the tabs is "So far", the summary the platform keeps of the conversation, then a row of buttons such as "Save as version", "Analyze the result" and "Tune parameters"; the contents are by tab:

Result

Backtest figures and the equity curve

Code

The strategy code, editable; click "Run backtest with this code"

Params

The strategy's parameters and risk settings; change values and click "Run backtest with these params"

Spec

How the platform read your strategy; you can change the backtest range here

Report

The analysis report (after "Analyze the result"), see Read the analysis report

Tuning

Tuning results (after "Tune parameters"), see Tuning

Factor

Factors written in the conversation (only when there is one)

Chart

Charts from data questions (only when there is one)

Runs

The conversation's runs; you can go back to one

Backtest settings

"Backtest settings" at the bottom of the "Params" tab takes the fee, slippage (in bps, 1 bp = 0.01%) and whether perpetual funding counts in the P&L. Leave a field empty for the platform default (the grey number in it). After "Save to this conversation", every later backtest, follow-up, tuning and analysis in the conversation uses them; the result's parameters show what was actually used. Spot markets settle no funding, so that setting is not there.

"Backtest settings" at the bottom of the "Params" tab

You can also enter fees on a strategy's page in the library, but they apply to that one run only and are not saved.

Starting capital is 100,000 USDT and gross exposure (leverage) 1× by default; say either in the conversation, such as "capital 20,000" or "3× leverage". Gross exposure can also be changed under the risk settings on the "Params" tab, then "Run backtest with these params". Perpetuals can be leveraged; if the gross exposure is over the cap, a note under the parameter says the most this market allows. Spot cannot be leveraged.

Memory and sources

Memory: state a lasting preference once and the Agent remembers it and follows it in later conversations, so you do not have to repeat it. For example:

When it remembers one, the reply shows "Remembered: …"; click "Undo" to take it back. All memories can be viewed and edited under "Settings → Memories", see Memories.

Sources: when a reply draws on your earlier strategies, runs, documents or factors, the sources are listed below it, numbered to match the [n] in the text; click one to check it.

Conversations and runs

The list on the left holds your conversations, each titled by its first message. Hover a row to "pin" or "archive" it; an archived conversation has to be restored before it opens. Conversations cannot be deleted or renamed.

The list can be filtered by recent activity or by state (running, unread, pinned, archived); the dot on a row means running, waiting for your approval, finished since you last looked, or failed since you last looked.

Every piece of work the Agent does (writing a strategy, a backtest, an analysis…) is a run, and each run has a detail page with each step, how long it took and the outcome. The list of all runs is under "Strategies" → any strategy → "Runs" → "All runs →".

Approvals waiting, the Agent's questions, and runs that failed or did not finish (except ones you stopped yourself) also show under "Notifications" at the top right.