Agentic Quant Workspace

Turn Your Agent into a Quant

Your Claude Code or Codex writes the strategy, backtests it and takes it live.

Free Download for Mac Free Download for Windows
GitHub GitHub

Free and open source · Download on a Mac or PC

Data, backtests and live trading, all free
  • Desktop app
  • Source code (Apache-2.0)
  • Public market data
  • Bring your own data sources
  • Backtest and guardrails
  • Live trading
Paid Blave AI·usage-based Cloud machine·monthly, data included
Supported markets
Crypto Taiwan stocks Taiwan index futures
Trade On Binance BingX OKX Gate.io Bybit Capital Futures
Data Sources FinMind TAIFEX TWSE TPEx

Every step it takes, you can see

01

Turn One Sentence Into a Strategy

Skip the code. Your AI writes it, pulls the data, and runs the backtest.

Chat with your AI: the user asks for the max drawdown on ETH Holder Concentration, the agent answers -40.55% with the current parameters (ENTRY_TH=1.0, EXIT_TH=0.5) and adds that the original thresholds drew down -51.41%, so the stable-region parameters cut about 11 points while raising Sharpe from 0.84 to 1.02; the input box below holds the next idea, "Exit everything when BTC breaks below its 20-day moving average"
02

Check That a Good Backtest Isn't Overfit

Overfitting: parameters that just happen to fit past data. MCPT shuffle test, Parameter Scan, OOS on unseen data. All three must pass.

Three checks on one backtest: MCPT p-value 0.024 on the Backtest tab, current settings inside the plateau on the Parameter Scan tab, and OOS efficiency 0.67 (just above the bar) on the Out-of-Sample tab
03

Deploy It and Let It Run

Paper trade first, no keys needed. Then connect your exchange and trade live.

Equity curve on the auto-trading overview (illustrative data): one month of equity, rising in the final stretch.
04

Get a Report on How It's Running

Ask for performance, positions and recent trades. Or put it on a schedule.

One-off report "Supertrend (SOL) - Running Status" on paper trading: a long position of 14.06 SOL, with entry price and unrealized PnL.

Partners

Community
Community
Business
Business

Questions?

The questions we get most; details are in the docs.

Read the Docs →

Not with your own Claude Code or Codex. You sign in only for Blave AI, the cloud machine or Blave data.

Three layers: the app and source code are free (Apache-2.0); your own AI runs on your own subscription, with no AI fee from Blave; what costs money is Blave AI (usage-based), the cloud machine (monthly) and Blave data (included with a machine, hourly otherwise).

The AI only researches and writes the strategy; orders are placed by deterministic code on a schedule, never by an LLM deciding long or short each round. Backtests come with an MCPT p-value, a parameter plateau and walk-forward, and fees cannot be set to 0.

On the desktop app they sit in the workspace .env on your computer; on a cloud machine they stay on your own machine; bound on the website, Blave stores them encrypted. Use trade-only keys with withdrawals disabled.

Binance, BingX, OKX, Gate.io and Bybit; Capital Futures (Taiwan index futures and Taiwan stocks) is currently supported on the cloud machine only.

Yes, for Windows 10 and 11 (x64). The installer isn't code-signed yet, so Windows warns on the first install: click “More info”, then “Run anyway”.

Yes. Describe your trading idea in plain language and your AI writes it into a strategy, backtests it, and runs it for you. You never touch code. No idea yet? Start with a ready-made official strategy.

You don't need one. The Strategy Library has ready-made, backtested strategies — pick one that fits and run it; once you know it, adjust the parameters and logic and make it your own.

The same workspace can move to your own cloud machine, reachable from the web, Telegram or SSH. See the cloud machine page.

Run a Backtest for Free

Free Download for Mac Free Download for Windows
GitHub GitHub

Free and open source · Download on a Mac or PC