Customer Support Agent

A tier-1 support agent that reads each ticket, replies on-brand when it can, and escalates the rest with a human-ready briefing — around the clock.

View in marketplace Open dashboard

The Customer Support Agent handles inbound support tickets: it parses the message, decides whether it can resolve it or needs a human, and writes a clear reply in your configured tone. If the ticket includes a sender address, it sends the reply automatically via send_email (using a connected Gmail/Outlook mailbox if available, otherwise the platform mailer); otherwise it drafts it for review. Tickets it shouldn't answer come back with a concise escalation briefing so a human can pick up with full context.

What it does

This agent is built to deflect repetitive tier-1 support load so your team only touches what truly needs them. Each run takes one incoming ticket, reads the customer's intent, and makes a resolve-or-escalate decision. When it resolves, it writes a friendly, formal, or concise reply (your choice) grounded in the product/company name you configure. When it escalates, it hands a human a short briefing instead of guessing.

Operationally it runs on three tools: web_fetch (to pull the readable text of a public https page when needed), send_email (to actually send the reply), and record_output (to log the deliverable). The send_email tool sends via the buyer's connected Gmail or Outlook mailbox if a token is available, and otherwise falls back to the platform mail service with a reply-to so customer replies still reach you. It only sends when the ticket carries a sender address; with no sender address it drafts the reply instead. Every run finishes by calling record_output exactly once, which writes a titled markdown deliverable (reply text, sent-or-drafted status, and any escalation) with a one-line summary shown in your dashboard feed. The run is bounded to 5 turns, keeping each ticket fast and predictable.

How it works

Your Business

Read the ticket

Parses the incoming customer message (pasted sample for test runs, or a live inbound message/webhook) and identifies the intent and any sender address.

Decide resolve vs. escalate

Judges whether it can answer confidently or whether the ticket needs a human (e.g. an edge-case refund or account action).

Ground the answer

When useful, calls web_fetch on a public https URL to pull readable page text before writing, alongside your configured product/company context.

Write the reply

Composes a clear reply in the configured tone — friendly, formal, or concise — referencing the specifics of the ticket.

Send or draft

If the ticket has a sender address, calls send_email (buyer's connected Gmail/Outlook, else platform mailer with reply-to); otherwise leaves it as a draft.

Escalate with a briefing

When it can't safely resolve, produces a concise human-ready briefing summarizing the issue and what's needed.

Record the outcome

Calls record_output once with the reply, sent-or-drafted status, and any escalation, writing a markdown deliverable to your dashboard feed.

Outcomes delivered

Setting it up — owner / admin

1
Deploy from the marketplace
Subscribe to the Customer Support Agent, then open its deployment page at /dashboard/agents/[id]. This is where you configure, test, and run it.
2
Connect a reply channel
For live sending, connect a mailbox under /dashboard/tools — Gmail or Outlook. send_email uses whichever is connected and otherwise falls back to the platform mailer with a reply-to. Slack/helpdesk connections are optional for escalations.
3
Fill the config fields
Set the three real fields: 'Reply tone' (Friendly / Formal / Concise), 'Product / company name' (what you support, used as context in every reply), and 'Incoming ticket (for test runs)' (paste a sample customer message).
4
Understand the trigger model
This agent is realtime/event-driven — it responds per inbound ticket rather than on a fixed schedule (there is no schedule field in its config). Wire your support inbox or chat/webhook to it, or invoke it on demand.
5
Do the first run
Paste a sample message into 'Incoming ticket' and click Run now. Confirm the tone, the resolve/escalate decision, and whether it sent or drafted. Connect Gmail/Outlook and rerun a ticket that includes a sender address to verify live send.
6
Review and iterate
Check the recorded deliverable in your dashboard feed, adjust tone or product context if replies miss the mark, then let it run against live tickets.

Using it day to day — your team

1
Submit a support request
Customers reach out as they normally do — through the support inbox or chat/webhook wired to the agent. No change to how they ask for help.
2
Get a fast reply
Resolvable tickets get an on-brand reply in seconds — sent straight to the customer's email when a mailbox is connected and the ticket has a sender address.
3
Hand off cleanly when needed
For tickets the agent escalates, a human agent receives a concise briefing with the issue and context, and continues the conversation without re-reading the whole thread.
4
See results in the dashboard
Owners and support staff find each run's deliverable in the /dashboard feed — the reply text, whether it was sent or drafted, and any escalation note — via the record_output entry.
5
Review drafts before sending
When the ticket has no sender address (or no mailbox is connected), staff read the drafted reply from the deliverable and send it manually.

Use cases

After-hours cover

Customers get instant, accurate replies even when your team is offline, so first-response times stay fast around the clock.

Refund/billing triage

Drafts or sends replies for common billing questions and escalates genuine edge cases with a summary a human can act on.

Knowledge-grounded answers

Answers product questions consistently using your configured product/company context, with web_fetch for live facts when needed.

Tier-1 deflection for SaaS

Resolves repetitive how-to and account questions automatically, leaving only the genuinely complex tickets for your team.

What to expect

A drafted or sent on-brand reply for every ticket the agent handles
A resolve-or-escalate decision per ticket, with a human-ready briefing on escalations
Faster first-response times, around the clock
A recorded markdown deliverable per run in the dashboard feed (reply + sent/drafted status + escalation)
Lower tier-1 support load so the team focuses on tickets that need them

Metrics to watch

First-response time on inbound tickets (target: seconds)
Auto-resolution / deflection rate vs. escalation rate
Share of replies sent live vs. left as drafts (depends on sender address and mailbox connection)
CSAT on agent-handled tickets
Escalation quality — how often a human can act on the briefing without reopening the thread

FAQ

Will it send emails on its own?

Only when the ticket includes a sender address. send_email uses your connected Gmail or Outlook mailbox if available, or falls back to the platform mailer with a reply-to. Without a sender address it drafts the reply for you to review and send.

How does escalation work?

When the agent decides a ticket shouldn't be auto-answered, it produces a concise human-ready briefing summarizing the issue and what's needed, instead of guessing — so a person can pick it up with full context.

How do I test it before going live?

On the deployment page (/dashboard/agents/[id]), paste a sample message into the 'Incoming ticket (for test runs)' field, set the tone and product name, and click Run now. You'll see the reply, the resolve/escalate decision, and whether it sent or drafted.

Can it match my brand and product?

Yes. Set 'Reply tone' (Friendly / Formal / Concise) and 'Product / company name' in the config; both are used as context on every reply. It can also pull readable text from a public https page via web_fetch when helpful.

Is this scheduled or realtime?

Realtime / event-driven — it responds per inbound ticket. Unlike the scheduled agents, its config has no run-frequency field; wire it to your support inbox or a chat/webhook, or invoke it on demand.