The product
Everything below works today, in the app, in English and Arabic.
Build
Code editor
Write Python agents in the browser, as one file or several, with packages from requirements.txt. backcrew handles the hosting, scaling and the model connection.
Builder assistant
Describe the agent in plain words, in English or Arabic. The assistant writes the code, suggests memory, tools, knowledge and a policy, and fixes its own mistakes before showing you.
Templates
Start from ready-made agents in the Marketplace, such as cloud operations or compliance evidence, and make them yours.
Models
Choose per agent: Claude Haiku 4.5, Claude Sonnet 5.5 or Amazon Nova 2 Lite. Change it any time; you're billed for the model each call really used.
Memory
Off, within a conversation, or across conversations, with how long it's kept set per agent.
Versions
Every save is a version. Compare any version with the current one and roll back in one step.
Connect
Tools
Any REST API, with an OpenAPI description or a generic one. API keys and OAuth client credentials are stored once and never shown again, not even to you.
Knowledge bases
PDF, Word, Excel, CSV, HTML and text, up to ten files at a time, and websites you've verified you own, re-crawled on a schedule. Answers cite the file or page.
Code and browser
Let an agent run code in a sandbox, or browse the web on an allowlist of sites you choose.
Control
Policies
Topics the agent stays off, with your own refusal message in each language; personal data such as emails, phone numbers and national ID numbers hidden; harmful content and prompt attacks refused; tool actions it may never take.
Test cases
Your questions with the answer you expect, as an exact phrase or a rubric judged by a model. Run them before each release and see what changed.
Quality checks
Score a sample of live answers against criteria you write, and get told when an agent's quality drops.
Code review
Every deploy is reviewed for security problems and its packages are checked against known malware before anything is built. It's free.
Ship
Website widget
One script tag. Allowed sites only, rate limits per widget, and visitors never see your balance.
WhatsApp, Telegram and Slack
Connect your own number, bot or Slack app. Each channel's conversations are kept apart and can be handed to a person.
Public API
API keys you create and revoke, answers inline or as async runs with signed callbacks to your server.
Schedules and webhooks
Run an agent every morning, or when another system calls its webhook, and get told when a run fails.
Inbox
If you turn transcripts on, every conversation is kept for the period you choose. Search, flag bad answers, assign, take over and reply.
Notifications
Handoffs, failed deploys and schedules, quality drops and low balance, in the app and by email, as each person chooses.
Deploying is free. You pay per request.
Sign up, build your first agent and see what each answer costs before you go live.