AI-assisted support replies for Intercom

The problem

Intercom's AI agent, Fin, costs $0.99 per resolution and operates autonomously — it reads the conversation and replies on its own. That works for simple FAQs, but for tricky billing disputes, frustrated customers, or edge cases that need a human touch, most teams keep Fin turned off or heavily restricted. The last thing you want is an AI making a promise your team can't keep.

So your agents handle the hard conversations manually. They copy conversations into ChatGPT to help draft responses, then paste them back. It works, but it's slow, the tone varies from agent to agent, and nothing about it scales.

How Revryte fixes this

Revryte is agent-assisted, not autonomous. The AI helps your agents write better replies faster, but the agent always reviews and sends. It works in Intercom in any browser on macOS.

1. Resolve with specifics — the agent types ;;resolve in the reply composer. The command template includes tags like <acknowledge their specific issue>, <explain what was done to fix it>, and <next steps if any>. The AI reads the conversation on screen — sees the customer reported broken webhook deliveries since switching to the new API version — and fills each tag. The agent reviews, adjusts if needed, and sends.
2. Quick note to polished message — the agent types checking with eng team, will update by eod then types ;;formal. It becomes: "I've escalated this to our engineering team and they're looking into it now. I'll follow up with an update by end of day today."
3. Specific thank-you after resolution — after closing a conversation, the agent types ;;ty. The AI reads the thread and writes a thank-you that references what the customer actually reported: "Thanks for flagging the webhook issue and for your patience while we traced it to the API version mismatch. That helped us identify it quickly."

Human in the loop

Fin replaces the agent. Revryte assists the agent. That's a fundamental difference. Your agents see the AI's draft, edit what needs editing, and decide what the customer reads. No automated replies going out that nobody reviewed. No AI hallucinating a refund policy you don't have. The agent is always the last checkpoint before a message is sent.

This matters most for the conversations where tone is everything: an upset customer, a sensitive billing issue, a bug report from your biggest account. Those are exactly the conversations where you want AI help with phrasing but human judgment on substance.

Shared commands for team consistency

Your CS lead creates the commands: ;;resolve, ;;escalate, ;;ty, ;;formal. Each one has a template with the right structure, tone, and tags. Every agent on the team triggers the same commands and gets replies in the same voice. New hires write like veterans from day one because the templates encode your team's best practices.

When your support voice needs to change — say you're updating your tone to be less formal — the lead updates the command templates once, and every agent's output reflects the change immediately.