Quick Start

Two ways to run it: the free, open-source desktop app with your own Claude Code or Codex, or a cloud machine that keeps going with your computer off.

Last updated 2026-09

Desktop App

1

Download the Mac or Windows app, install it and open it. Free and open source (Apache-2.0); no sign-up needed first.

  • Mac: macOS 13 or later, one installer for Apple silicon and Intel. Open the .dmg and drag Blave into Applications.
  • Windows: Windows 10 or 11 (x64). It installs under your own user account; no administrator rights needed. Windows on ARM isn't tested yet.

Windows Warns on the First Install

The installer isn't code-signed yet (we've applied for a certificate), so Windows can't identify the publisher and stops it once. The source and the build are public on GitHub; every release is built by a public GitHub Actions workflow.

  1. If your browser says the file isn't commonly downloaded, keep it. In Edge: “…” → Keep → Show more → Keep anyway.
  2. Open the installer. When you see “Windows protected your PC”, click “More info”.
  3. Check that the file name is Blave-Setup.exe, then click “Run anyway”. Blave installs and opens on its own.
  4. The first launch sets up its runtime and takes a few minutes; after that it opens right away.

Don't see “Run anyway”? A work PC's policy, or Smart App Control on Windows 11, can block apps that aren't signed yet; neither can be bypassed from the dialog. You can use a cloud machine for now.

2

Pick your AI: bring your own Claude Code or Codex — it runs on your own subscription, no Blave account needed — or sign in and use Blave AI, billed by usage.

3

Describe your trading idea in one plain sentence. It writes the strategy, runs the backtest, and shows the results right in the workspace.

4

Happy with the backtest? Connect your exchange keys and the strategy trades live; the keys sit in the workspace .env on your computer.

Keep It Running With Your Computer Off: Launch a Cloud Machine

1

Log in and add a card on the launch page. A first card is charged only a 10 TWD verification (it goes into your balance) and comes with 14 days of waived server fee plus 100 TWD of AI credit; the launch button unlocks once the card is on file. To top up or turn auto top-up off later, go to Usage & Billing.

Credit card on the launch page: select an amount and bind a card
2

Click Launch Server. If you like, open Advanced settings first to pick the operating system (Linux by default, or Windows) and server tier.

3

Wait a few minutes while your server installs — the status badge goes from Pending to Installing to Running.

4

Click Enter Workspace and start chatting: ask for market data, have the agent build and backtest a strategy — everything runs in the web workspace.

Server card on the launch page with the Enter Workspace button The web workspace right after entering: strategy sidebar and starter actions
5

Optional — prefer chatting in Telegram? Follow Connect Telegram to create your own bot and pair it.

Windows server? You can also work on the full desktop over Remote Desktop — see Remote Desktop.

Agent standing rules

Some things you only want to say once: “always include fees in a backtest”, “default to USDT-margined perps unless I say otherwise”. Those are standing rules — once written down, the agent reads them every turn, across the cloud machine's Telegram, web workspace and SSH, and they survive a new conversation. Two ways to write one:

  • Say it in the chat. Tell the agent “from now on every strategy needs a stop loss” and it records that as a rule.
  • Manage them in Settings. Workspace, More › Settings › Agent rules: read, edit and delete them there.

The limit is 10 rules, 150 characters each. When it is full it stops and tells you — nothing old gets dropped for you. Delete one yourself to make room; we don't decide which of your rules to sacrifice.

Rules are defaults, not hard limits: safety and halt rules (HALT, close-all, risk caps) rank higher and always win. When a rule clashes with the strategy logic — a portfolio strategy has no per-trade stop, for example — the agent says so and asks you. It won't silently comply, and it won't silently ignore you.