Built around one rule: a person approves anything that leaves your business.
Every capability below is live in the product today. Where something is not yet proven end to end, it says so.
Email assistance (Gmail)
Inbox monitoring
Laboryn is notified of new messages in the connected inbox and classifies each one (question, order status, complaint, and so on).
Draft replies in your Gmail
A suggested reply is written into your own Gmail Drafts, on the right thread, addressed to the original sender — chosen from the message headers, never from text inside the email.
Review queue
Anything sensitive or uncertain — refunds, payments, legal, low confidence — is held for a human in the Decision Center. No automatic send.
Optional auto-send
A workspace can opt in to automatic sending for clearly safe categories only. It is off by default and capped.
Order status (Shopify)
Read-only
Laboryn requests read_orders and read_customers only. It never writes to your store.
Local answers
Order-status replies are built from your own synced order data. Anything that looks like a refund or change of order still goes to a human.
Knowledge & missions
Approved business facts
You curate the answers Laboryn is allowed to use. For general questions it will only state facts from that approved set, and it says so when it lacks one.
Missions
Ask for a plan, an audit or a review; Laboryn routes it to the right internal agents and keeps the work durable.
Data handling
What is sent to the AI
The plain-text message body and your approved facts. Not attachments, not raw MIME, not headers beyond a minimal set, not tokens or cookies.
What Laboryn keeps
Message bodies are not persisted (retention zero). Short encrypted "decision snapshots" (sender, subject, a summary, the drafted reply) auto-delete after 72 hours, or immediately on disconnect.
At rest
Connector tokens and other secrets are stored with per-record envelope encryption bound to your workspace.
Not yet claimed
A managed KMS/HSM, a completed third-party security assessment, and the live real-provider end-to-end runs are still pending. We do not claim them.