Quickstart

This takes about ten minutes, most of it waiting for the first deploy.

Open backcrew and sign in with the email you were invited with. You land on the Dashboard: live agents, balance, spending and recent usage.

2. Build an agent

Open Agents in the sidebar and click New agent. Choose how to start:

Start from a template - opens the Marketplace. Click Install on a template, such as Arabic Customer Service, choose your tools if it asks for any, and it deploys. You don’t see or write any code.
Describe it - pick a model (its price is shown next to it) and a region (where it runs), then say what it should do in plain words, in English or Arabic. For example: “Answer customer questions about returns and shipping, in Arabic and English.” Click Generate code. Before anything is created, backcrew shows what the agent will do, what it will use and what it won’t do, plus a suggested policy. Click Create this agent to create it (and the policy, if you keep it) and deploy it. The code stays folded away under Code unless you open it.
Write code - the full Builder with the editor up front. The code is a normal Python project: agent.py defines agent = Agent(...) with the Strands Agents SDK, and requirements.txt lists any extra packages.

3. Deploy

A template deploys when you install it. Otherwise, click Deploy. backcrew builds the agent’s image and creates its runtime and everything it uses. The status moves from Building to Deployed - usually a few minutes.

4. Try it

Open the agent’s page and send a message in the Invoke panel. Each answer shows the tokens used and what it cost, in credits and dollars.

A long answer is fine: the request keeps running while the panel waits for it, for up to a few minutes.

Open How it answered under an answer to see what the agent did: each tool it called, with its arguments (passwords and keys hidden), whether it worked and roughly how long it took; each document passage it found, with the file or page it came from; and whether your policy blocked or changed the reply, by the policy’s name. An agent deployed before this feature shows it after its next deploy.

When it doesn’t answer

The panel says why, for example that the agent’s code raised an error, or that the workspace is out of credits. Click Show the agent’s logs to see what its code printed around that request, including the full error and the line it came from. Logs can take up to a minute to arrive, so if there’s nothing yet, click Refresh logs. Every request, answered or not, is also listed on the agent’s Activity page.

If the logs show an error in the agent’s code, click Fix with AI. It opens the editor with that error, and the Builder assistant proposes a fix for you to review before saving. It’s charged like any assistant request.