Keel Docs
Getting Started

Your First Backtest (Web App)

Walkthrough of building and backtesting a strategy in the Keel web app. CLI-first users should follow Getting Started instead.

Your First Backtest (Web App)

The web app at app.usekeel.io is the visual alternative to the CLI. Same backend, same strategies, same backtests — different surface. In the app you build by talking to the AI agent or by editing the strategy directly on a canvas; there is no drag-and-drop node graph to wire up.

CLI-first users — read Getting Started instead. That's the v1 install path: pipx install keel-trade + keel auth login + your agent has full access. This page is for users who'd rather work in the browser.

Prerequisites

Step 1: Get a Strategy

Three ways, pick one:

  1. Ask the agent. Every strategy has a chat panel. Describe what you want — "build a momentum strategy on the top 25 perps" — and the agent composes the pipeline for you, explains it, and keeps editing it as you refine the idea.
  2. Fork from the Library. Open Library in the app, pick a published strategy, and fork it into your account. You get the exact code behind the published run — re-run it yourself to verify.
  3. Start blank. From Strategies, click New strategy and build in the editor.

Step 2: Understand the Editor

The editor has two views, toggleable in the toolbar:

  • Visual — the pipeline as an ordered stack of components (data → signals → sizing → execution). Components run top to bottom; there are no edges to connect.
  • Code — the same strategy as Keel DSL source, editable directly.

Both views validate continuously. Errors and warnings appear inline — fix anything red before running. The agent can fix validation issues for you if you ask.

Step 3: Run the Backtest

Click Run Backtest in the toolbar (or just ask the agent to run one). Pick the strategy version, set the date range, and run. Backtests execute server-side against Keel's cached Hyperliquid price + funding history — you can keep working while it runs.

Step 4: Read the Results

The backtest report shows an equity curve plus the core metrics:

MetricDescription
SharpeRisk-adjusted return — return per unit of volatility
ReturnPercentage gain/loss over the period
Max drawdownLargest peak-to-trough decline
Win rateClosed trades that made money
TradesNumber of trades placed

A run that placed no trades says so explicitly — the numbers on such a run describe idle capital, not a working strategy. Ask the agent "why didn't this trade?" and it will read the run's own warnings.

What's Next?

  • Switch to the CLI / agent flow for keyboard-driven and AI-driven iteration
  • Agent Setup — wire your strategy into Claude Code, Cursor, etc.
  • Deploy live once you're satisfied with the results — see "Going live" in Getting Started