Clear the routine queue while uncertain cases reach a person.
Summarize and classify new support requests, ground suggested responses in approved knowledge, and escalate anything outside the playbook.
How this composition works
This workflow creates a consistent first pass across email, workspace discussion, and community channels. The agent identifies the issue and missing evidence, checks the approved knowledge source, and prepares a draft. Sensitive, low-confidence, or consequential cases are routed to a human.
Collect new requests
Read new messages from the approved support inbox and channels.
Classify and route
Identify issue type, urgency, ownership, and missing evidence.
Ground the draft
Use only approved product and policy knowledge when proposing a response.
Watch community reports
Include relevant authorized community issues without posting autonomously.
Connected systems
These are the real product connectors composed for this example, each scoped to the credential permissions your workspace approves.
Gmail
Search and read Gmail messages and threads (read-only).
Explore connectorSlack
Read channels and messages, post messages, and upload files as your Slack bot.
Explore connectorNotion
Read, search, and update Notion pages and databases.
Explore connectorDiscord
Read and post messages in Discord servers, and read profiles and server details.
Explore connectorOptional skills
Where relevant, skills add reusable method and verification. The connectors and agent instructions remain the primary building blocks.
Try asking
More examples for Customer support
Build a governed workflow like support queue triage.
Tell us what your team repeats today. We will map the systems, approval boundaries, and first governed run with you.