No secret pasting
Lead forms and trial shifts ask for the minimum context needed. Buyers are told not to paste secrets, credentials or private account access.
Trust
AI robots start with submitted inputs, not broad account access. Every run leaves a receipt. Risky, public or customer-facing use needs human approval first.
Lead forms and trial shifts ask for the minimum context needed. Buyers are told not to paste secrets, credentials or private account access.
A person checks AI robot fit, source safety, output type, approval point and receipt expectations before larger work is discussed.
Work should leave a readable record of order, steps, checks, output, evidence, limits and next action.
Production credentials, customer-facing publishing and risky use require explicit human approval.
Example receipts and demo data are not represented as customer outcomes.
DexCode and Bex run as separate apps with their own auth, product data and deployment boundaries.
Data boundaries
AI robots work from the minimum approved context for the job. The boundary is the work order, the approved integrations, the credential references, the permission policy and the receipt a human can review.
A robot receives only the source material, integration access and instructions needed for the approved job.
The app stores credential reference names and policy state, not raw secret values in page copy, receipts or prompts.
Runs produce receipts with inputs used, steps taken, checks run, limits found and the next approval decision.
Project-level settings can override organization defaults, including Dex and Bex cloud credential references.
Retention and redaction
The current policy is a manual review boundary, not automatic deletion. A person checks legal, customer and product evidence needs before records are deleted or redacted.
Private lead records are reviewed after 18 months. Keep only active customer conversations, current sales follow-up or records with a specific operational reason.
Private trial work orders, private receipts, runs and run events are reviewed after 12 months, with prompt-like fields redacted when the receipt is no longer needed for a buyer conversation.
Public example receipts are product evidence. They are edited deliberately in product data, not swept by private-record redaction tasks.
Budgets, permissions, approval points, retention and credential references define what a robot can do before work starts.
Runs capture events, checks, artifacts, cost signals, failures and recovery notes so the outcome can be audited.
Risky, customer-visible, production, spending or external-message actions require explicit human approval.
Browse AI robot resumes, review example receipts on the pricing page or start from a small work-order template.