Every ticket sorted
before a human opens it.
Support inboxes drown teams. Half the messages are duplicate, a quarter are spam, and the urgent ones get buried by accident. A Loomal triage agent reads each new message in its own inbox, classifies it, applies labels, and routes the urgent ones — so humans only see what actually needs them.
API Primitives used
mail_list_messagesRead every new message
Agent polls its inbox continuously. extractedText strips signatures and quoted history so the LLM sees only the customer's actual problem.
mail_update_labelsTag and prioritize
Apply structured labels — severity, product area, sentiment — directly to the message so downstream tooling routes correctly.
mail_replyAcknowledge instantly
Send a real, threaded acknowledgement so the customer knows they're heard while the right human picks up the case.
Support queues hide the urgent issues.
The hardest job in support isn't answering tickets — it's finding the right ticket to answer next. Every shared inbox accumulates noise: feature requests, marketing replies, accidental forwards, automated bounces. The 1% of messages that represent a customer in crisis sit next to the 99% that don't, and the agent on shift has to sort them by hand.
Triage automation has traditionally meant rigid rule engines: if subject contains 'urgent', flag it. Customers learned that game. The real fix is reading every message with a model that actually understands intent — and acting on the read directly inside the inbox, not in a separate ticketing platform that adds another sync layer.
How to build it.
mail_list_messagesRead the inbound
Agent fetches new messages on a tight poll cycle, parses extractedText, and runs each through a classifier prompt.
mail_update_labelsApply labels and route
Severity, product area, customer tier, and sentiment are written back as labels. Routing rules consume those labels in real time.
mail_replyAcknowledge in-thread
Customer gets a real reply from the support identity confirming receipt and setting expectations — within seconds of sending.
Example prompt
“Watch support@. For each new message classify it as bug / question / billing / feature / spam, label it with the matching tag and severity, and reply to anything tagged 'urgent' confirming we're on it.”
What support teams build.
Severity classification
Agent reads each message and assigns P1–P4. P1 messages alert the on-call engineer immediately with full context attached.
Product-area routing
Each message gets a product label. The right team's queue picks it up automatically with no manual sorting.
Duplicate detection
Agent recognizes multi-message threads from the same customer and merges them so reps see one conversation, not five.
Spam filtering
Agent flags non-customer messages — vendors, sales pitches, automation bounces — out of the human queue entirely.
Sentiment escalation
Messages with frustrated or churn-risk sentiment get bumped in priority and routed to a senior rep automatically.
Why triage belongs in the inbox.
Most triage tools live outside email. They sync messages into a ticketing system, run rules, and write back. Every sync is a chance for state to drift, and every external system is another integration to maintain. A Loomal agent skips the layer entirely — it reads, labels, and replies inside the same inbox the customer wrote to.
The result is faster routing, fewer dropped messages, and a complete audit trail tied to the support identity. Compliance teams see exactly which messages were classified, by which agent, on whose authority — without piecing together logs from three vendors.
Audited classifications
Every label change is logged with the agent identity and timestamp — defensible if a customer disputes how their issue was handled.
Identity-scoped inbox
Triage agent only sees the support inbox it owns. Other teams' mail stays isolated.
Revoke on the fly
If the agent misclassifies, revoke the identity instantly to halt every action — no cleanup scripts.
Stop sorting tickets by hand.
Real inbox triage, real-time labels, fully autonomous.