## Summary

AppReply MAX is our most capable review response agent yet. It combines live URL-based knowledge with Agentic Memory, a self-improving layer that scores 95%+ on LongMemEval, so replies stay accurate and on brand even when review volume spikes.

For mobile app and game teams, review response quality usually breaks at the exact moment volume goes up. That's after a release, a pricing change, a live ops event, or a support spike, when replies need to be most accurate and your team has the least time to write them.

AppReply MAX is our answer to that moment. It's the most capable review response agent we've shipped, combining live URL-based knowledge with a self-improving memory layer that scores 95%+ factual accuracy on long-memory benchmarks. The goal is simple: keep replies accurate, on brand, and up to date even when volume triples overnight.

### 1. Link URLs as context. Skip the upload.

Knowledge bases used to mean uploading files every time you shipped. That worked when your docs changed once a quarter. It does not work when your help center, changelog, pricing page, and FAQ move every week.

With AppReply MAX, you paste a URL. AppReply crawls the page, indexes it, and keeps it in sync on a cadence you choose.

Live URL sources live alongside any documents you still want to upload manually, so teams can move existing knowledge over at their own pace.

The practical effect: one less release-day chore, and replies that stay aligned with the latest version of your help center, changelog, pricing, and FAQ pages. **Available now on the Pro plan.**

### 2. Agentic Memory, with 95% factual accuracy

AppReply has always been good at finding relevant past responses. The harder problem is memory: remembering which replies actually worked, which edits your team made on the way to "send," and which tone fits your product when reviews get emotional.

That is what Agentic Memory is built for.

On LongMemEval, a public long-memory benchmark, Agentic Memory scores 95%+. Base models land around 62%. Most vector-memory systems land between 70% and 85%. For app and game teams, that gap shows up in a concrete way: more drafts you can approve on first read, and fewer replies you have to rewrite.

When Agentic Memory is enabled on your account:

- Your edits train the system over time, so repeated mistakes stop repeating.
- The agent starts recognizing patterns across reviews, including recurring complaints, fixes that resolved them, and high-stakes situations that need a lighter touch.
- Replies improve based on what has worked for _your_ app, not generic support language.

Under the hood, Agentic Memory builds a live map of your app's recurring issues, fixes, and language. The edges between nodes are what let new replies sound like someone who already knows your product.

After a few days, you'll edit less. After a few weeks, AppReply starts sounding like someone who already knows your product, your players, and your quality bar. That's the point at which 90%+ automation rates become realistic rather than aspirational.

One useful side effect: your release notes are now stored in Agentic Memory automatically. The response agent stays current with shipped changes without anyone having to paste notes into a knowledge base.

### How to turn it on

Agentic Memory is configured per account while we roll it out. If you'd like it enabled for your app portfolio, message us from the Chat panel in your dashboard and we'll set it up.

Live URL sources are available to all Pro accounts today. You can add your first source from **Knowledge base** in the sidebar.

### A small change to reply caps

As we roll out the new retrieval and memory layers, monthly reply caps on a few plans will come down slightly to protect reply quality. They remain among the most generous in the category: 1,500 on Standard, and 5,000 on Pro with learning enabled.

If your team is running close to the new limits, reach out and we'll find the right fit.