Product Hunt Picks · Nov 5: Giving Agents Real Devices and Memory

Found some interesting new products on Product Hunt today. Number one is Scalekit, 400+ upvotes, over a hundred comments.

Yeah, I checked that out. In one sentence: it gives production agents a unified MCP gateway.

An MCP gateway, sounds like developer infrastructure. What pain point does it actually solve?

Right now the most annoying part of hooking agents up to tools isn't lack of capability — it's that permissions and auth are way too fragmented.

Every app has its own auth setup. An agent connecting to a dozen tools — just managing tokens is a headache and a half.

So it pulls all that into one layer?

Exactly. Through one gateway, agents get access to 30,000+ tools across 600+ apps.

The key thing is, every action runs as the user who initiated it, and permissions never exceed what they've granted.

I like that design. It basically locks the agent's boundaries to whatever's authorized.

There's also a Virtual MCP Server that only exposes the tools each agent needs for its specific job.

Least privilege as a product feature. What about security?

Every call gets logged. For teams going to production, that's pretty much non-negotiable.

The free tier gives you 10,000 calls a month — plenty for pilot teams.

To sum up: Scalekit turns agent tool integration and authorization from dirty work into a standard layer.

Alright, next one. Glasshand — great title: giving agents a real iPhone and Mac.

Literally. The hardware sits in their data center, and an agent can rent one with a single prompt.

Wait, so you hook it up to Claude Code or Codex and they operate real devices directly?

Yep. Post to TikTok, call an Uber, compile and test iOS apps — anything. Just like a person would.

I totally get this pain point. So many things you just can't test in a simulator — you need real hardware.

And you get exclusive access during your rental. When you're done, it wipes the data and factory resets.

So how's this different from cloud phones or device farms?

The key is their customer is the agent, not a human. The interface and interactions are designed for agents.

Got it. Treating a device as just another action an agent can call — that abstraction is clever.

And then there's the third one, Living Brain. A memory layer for agentic AI.

This tackles an old problem. Every conversation has knowledge worth keeping, but it's trapped in each individual chat.

It extracts that and links it to your existing knowledge, turning it into continuously evolving long-term memory.

The cross-model part is pretty key. Can it connect via API, SDK, and MCP?

Yeah. Memory carries over between ChatGPT, Claude, Gemini, Cursor and the rest.

So every new conversation picks up where the last one left off, instead of starting from scratch.

But here's the catch. Can memory format and recall quality stay consistent across models?

That's probably why it got fewer votes than the first two. Concept is right, but execution is still being proven.

You hit the nail on the head. Everyone wants to build a memory layer — the hard part is interoperability and quality.

So to wrap up — today's three products have an obvious common thread.

They're all agent infrastructure. Gateway handles access, devices handle execution, memory handles context.

Right. People aren't competing on models anymore — they're competing on the agent foundation. My pick for today is just one: Scalekit.

Agreed. Authorization and logging are things you can't avoid when agents go to production. It tackles the pain point first.