Skip to main content

Agent 920

Agent 920 is an AI analyst for your time based (920) strategies. Instead of changing a stop loss, re-running a backtest, noting the result and repeating, you describe what you want to test and the agent runs it, compares the runs, and tells you what changed.

It works on a strategy you already have. Agent 920 doesn't build strategies from scratch - that's what Signals AI does for indicator based strategies.

Access and cost

Agent 920 needs a Limited or Unlimited Backtest plan - it isn't available on the Free Backtest plan. Every reply spends shared AI tokens.

Where to find it​

Agent 920 lives on the Backtest page. Look for the Ask anything related to your strategy backtest… bar along the bottom of the page, with the Agent 920 button at its right hand end.

It needs a strategy first​

Agent 920 works on a strategy that's loaded in context, so the launcher only becomes usable once you have one. Either:

  • Run a backtest at least once on the strategy you're building, or
  • Open a saved strategy from the Saved Strategies panel on the left.

Until one of those has happened there's nothing for the agent to analyse, and it will ask you to load a strategy before it runs anything.

The chat panel​

Once open, the panel gives you:

ControlWhat it does
Model pickerChoose which AI model answers - see Choosing a model
AI tokensYour current balance, with Buy tokens beside it
Suggested promptsOne-click starters such as "Test SL of 20%, increment by 10%, up to 100% at leg level"
New ChatStart fresh, clearing the current conversation
HistoryReopen earlier conversations
Voice inputSpeak your request instead of typing
Check its work

Agent 920 can make mistakes. Treat its analysis as a starting point and verify the numbers before you trade on them.

What you can ask it to do​

Run backtests​

Ask for a specific configuration and the agent runs it:

"Backtest this with a 30% stop loss."

"Run it again with entry at 9:30 instead."

Each run tests one configuration. To sweep a parameter, ask for the variants you want - the agent runs the first, confirms the setup is right, then runs the rest together.

Relative date ranges work as you'd expect - "backtest the last six months" or "how did it do last year?" - resolved against the current date in IST.

Compare runs side by side​

"Compare those three runs."

The agent renders a comparison card of the runs with a short label for each (SL 30%, SL 50%, Entry 9:30) and a plain summary of which performed best and why - highest profit, best win rate, and so on.

You can also attach specific backtests from the comparison view and ask the agent to analyse just those, rather than everything you've run in the session.

Filter results by weekday​

"Show me only the Monday results."

"How does it do on Tuesdays and Thursdays?"

"What happens if I drop Friday trades?"

Every performance metric is recalculated on the filtered set - P&L, drawdown, win rate, expectancy - not just filtered for display. Weekend days are supported for strategies that trade markets open on those days.

Filter results by DTE​

"Show me the 0 DTE trades."

"Filter for 1 to 2 DTE."

The same recalculation applies. Longer dated DTEs are supported, not just the 0-5 range.

Apply real world costs​

"Add 1% slippage."

"Apply ₹20 per order and show me the net."

"What does this look like with slippage, brokerage and taxes?"

The agent reports the costs it applied, gross P&L, a breakdown of total costs, net P&L, and every summary metric recalculated. Costs can be combined in one go.

If you ask for brokerage without saying how it's charged, the agent asks whether it's per order or per lot, and for the amount - it won't guess.

Working with legs​

Your strategy's legs are numbered from 0, in the order they appear in the strategy. Refer to them that way:

"Change the stop loss on leg 1 to 40%."

Where a parameter exists at both overall and leg level - stop loss, target profit, lot size - the agent asks which one you mean before changing anything. It won't assume.

What it's careful about​

  • Over optimisation. If you sweep a parameter until one value looks excellent, the agent will say so. A configuration that only works on one historical window is curve fitting, not an edge.
  • Balanced reads. Analysis covers weaknesses as well as strengths - a strategy with strong returns and a brutal drawdown gets described that way.
  • Costs. Slippage, brokerage and taxes are treated as part of the picture, not an afterthought.

Markets it covers​

Agent 920 works across every segment on the Backtest page:

SegmentInstruments
Weekly & Monthly ExpiriesNIFTY, SENSEX
Monthly Only ExpiryMIDCPNIFTY, BANKNIFTY, FINNIFTY, BANKEX
Stocks - Cash / F&OAll NIFTY 500 stocks
CryptoDelta Exchange and CoinSwitch

For index strategies it assumes NIFTY unless you say otherwise, and BANKNIFTY for bank focused ones. Amounts are shown in rupees.

On crypto, remember that markets trade on weekends - so weekday filters cover all seven days, not just Monday to Friday.

Getting the most out of it​

  • Load a strategy first. Run a backtest or open a saved strategy before you start - the agent needs one in context.
  • Change one thing at a time. Comparisons are far easier to read when each run differs in a single parameter.
  • Say what "better" means to you. Highest P&L, smallest drawdown and best win rate rarely point at the same configuration.

If something goes wrong​

If a backtest or an analysis fails, the agent explains what happened. For anything it can't resolve, reach the team at t.me/algotest_in.