Unbreak does what you set it up to do. Everything below is a setting you decide, and you can change it at any time.
$ claude mcp add unbreak$ codex mcp add unbreakPasswords, card numbers, and any field you mark are redacted before your support agent reads them.
Your support agent pulls only what it needs into the conversation — nothing else leaves the browser.
Every change is a draft until you publish it — test it in a private chat, and roll back in one click.
Paste one script tag into your site — that's the same widget you saw in the hero. No SDK, no build step.
Credits pay for your AI agents' work, by what each AI call costs. The smallest reply is one credit; a reply that reads a long conversation or runs several steps uses more. Follow-ups use credits too, and so do the assistant and agent tests beyond their free monthly usage. Replies from your human teammates are free and unlimited, and you start with 500 free credits.
Your AI agents pause — nothing breaks. Conversations keep arriving in your inbox, your team can answer everything directly, and agents resume when your plan renews or you upgrade.
Browser debugging context is coming soon. The planned release includes console errors, failed-request metadata, session replay, and on-device masking controls.