Answers from your facts
You write the business context once — offering, hours, shipping, policies, tone. The agent answers from it and refuses to invent what it doesn't know.
🎧 Customer Support Agent
An AI customer service agent for your DMs and comments
Customers ask the same questions in your DMs every day — hours, shipping, sizing, “where’s my order?”. This agent reads each conversation, answers from your real business context, and drafts the reply for your team to send. When you trust it, flip it to auto.
Draft-first by default · pauses when a human takes over · every run logged
What it does
Every conversation, handled the way you would
Looks up live data via MCP
Connect your order system, booking tool, or help docs as MCP servers and the agent checks real order status and availability mid-conversation.
Matches your voice
It studies how your team actually replies — greetings, emoji, sign-offs — so drafts sound like you, not like a bot template.
Knows when to escalate
Angry customer? Refund decision? It marks the thread urgent, leaves a note with what it found, and gets out of the way.
Chooses its scope
Comments only, DMs only, or both. Public comment replies stay appropriate for an audience; DMs can go into account detail.
Learns from your edits
Every time you correct a draft before sending, it saves the lesson — visible, editable, and applied to the next conversation.
Trust dial
Start on draft. Earn your way to auto.
Autonomy is a dial, not a switch. Most teams run the agent in draft mode for a couple of weeks, watch the quality, then switch busy channels to auto — with the guardrails still on.
Draft mode (default)
The agent attaches its suggested reply to the conversation as an internal note. Your team reviews, edits, sends — and the agent learns from every edit.
Auto mode (opt-in)
The agent replies on its own, capped per conversation per hour, clearly attributed, and automatically silent for hours whenever a teammate joins in.
Per-conversation pause
One click on any thread mutes the agent there — for the tricky customer, the influencer negotiation, the conversation you just want to own.
When-away mode
Prefer to answer yourself? Set the agent to only step in when you haven't replied on that channel recently.
A real run, step by step
1 · Customer writes
“Hey, my order arrived today but the bag was open and half the beans spilled 😞 Order #2481.”
2 · Agent reads the thread
Full history + your business context + how your team usually handles damage claims.
3 · Agent checks your systems
lookup_order(#2481) via your connected MCP server → shipped Monday, batch 240-C.
4 · Agent drafts
Apology in your voice, replacement offer per your policy, asks for a photo of the batch code.
5 · Your team sends
One tweak to the wording. The agent saves the correction and applies it next time.
Works well with
🎯 Sales & Lead Agent
When a support conversation turns into buying intent, the sales agent captures and qualifies the lead.
🛡️ Community Moderator
Keeps spam and scams out of your comments so the support agent only spends credits on real customers.
⚡ AI inbox triage automation
Classify every new conversation by sentiment and intent, alert Slack on complaints, and tag the rest.
Customer support agent — FAQ
How does the AI customer service agent know my business?
During setup you give it your business context in plain language — what you sell, hours, shipping times, return policy, tone. For live facts it can't know (order status, stock, bookings), you can connect your systems as MCP servers and the agent looks the answer up during the conversation. Everything it uses is visible in its run log.
Does it reply automatically or does my team stay in the loop?
Your choice, per agent. In draft mode (the default) it attaches a suggested reply to the conversation as an internal note — your team reviews, edits, and sends. In auto mode it replies on its own, with an hourly rate limit per conversation, and automatically stands down for hours whenever a teammate replies. You can also pause it on any single conversation with one click.
Can it handle comments as well as DMs?
Yes — you choose its scope during setup: comments only, DMs only, or both. Comment replies stay public-appropriate; DM replies can go deeper into account-specific detail.
What happens when it doesn't know the answer?
It says so and escalates. The agent is instructed never to invent prices, policies, or promises — when a conversation needs a human (an angry customer, a refund decision, something outside its context), it marks the thread urgent and leaves a note explaining what it found.
Does it get better over time?
Yes. When your team edits a drafted reply before sending, the agent compares your version against its own and saves the lesson — tone corrections, policy details, phrasing you prefer. Its memory is capped, scoped to the channel, and fully visible: you can read, edit, or delete anything it has learned.
Which platforms does it work on today?
It rides on Nimply's unified inbox. YouTube comments are available now; Instagram and Facebook DMs and comments arrive as Meta's app review completes; X (Twitter) DMs follow their platform approval. The agent itself is platform-agnostic — as each network unlocks, it just starts working there.
Stop copy-pasting the same answers into DMs
Hire the support agent free, give it your business context, and review its first drafts today.