ment is back [remotely]

The gap + where we’re at

We owe you an update.

I dropped off the face of the earth after shipping 1.0. Life happened fast (house stuff, a move, a growing family, the whole blur), and ment went quiet.

But we’re back and shipping again.

ment is still what it set out to be: local-first AI chat that stays private on your devices. Easy model installs, quick testing, and customizable assistant profiles. Simple app. Serious power.

The big next release: remote inference

Midway through developing the first update, it became clear we needed a pivot: do one thing well now that unlocks faster releases later. Remote Inference is the first step toward making ment multi-platform. Remote Inference currently lets you take multiple Macs running ment on the same network and turn one into a host & another into a client.

If you’ve got a Mac Studio or a beefy desktop that can chew through models, but you prefer to read your tokens from a MacBook Air, this update is for you.

How it works (host vs client)

  • Host Mac: you enable hosting in ment, pick which models are available, and the host does inference. Weights live on the host.
  • Client Mac: after pairing with the host for the first time you simply reconnect and chat like normal; ment streams responses back & forth over your LAN.

Why this matters

  • It decouples “what you use” from “what you run”: a quiet/light Mac can be your daily driver while the powerful machine does the heavy lifting.
  • It’s the first step toward making ment feel like an on-device tool that still scales with your own hardware (without turning you into a full-time ops person).

What’s included in this update (ship scope)

This release is about bringing increased stability to ment, and taking the security/privacy of Remote Inference just as seriously as we do any other conversation or message we store.

  • Secure device pairing for Remote Inference (so you know exactly what you’re connecting to).
  • Resilient discovery + host status (so the client can find the host and you can tell when it’s actually ready).
  • Trustworthy remote model selection (so the client can’t be tricked into thinking it’s using one model while the host is using another).
  • A fair and responsive host (so remote sessions don’t turn into runaway work or a janky experience).
  • Model integrity checks for downloads (basic “verify what we downloaded” safety: hash/size and provenance guardrails).
  • Security baseline hardening (hardened runtime / trimming entitlements where we can, and tightening key handling).
  • LAN-only (no cloud required) with end-to-end encryption & authentication after pairing.

Under the hood, remote already has a secure transport and a bounded, versioned protocol. We’re building the rest of the product-quality experience around that.

Readdressing old promises

If you read the June 2025 post, we threw around “over the next few weeks” a bit too freely. Reality check: we shipped 1.0, then life pulled focus, and we didn’t earn the right to keep making timeline promises like that.

So here’s the reset:

  • iOS / Vision / iPad: still on the table, but it’s not “just over the horizon” until we pave our way there.
  • Performance tuning: still a priority. Remote Inference makes performance even more important, and we’ll continually take swings at it as we go.
  • Image generation: further down the road than multi-platform ment, but once the remote foundation is stable (and once we can do it in a way that doesn’t turn setup into a mess), this is expected to follow on the roadmap.

Timing + how we’ll communicate

The update is coming soon, but isn’t imminent (“next few weeks”). I won’t be pulling the trigger until I’ve done my best to break everything and see that:

  • Pairing is straightforward and secure.
  • Discovery works reliably on a normal home LAN (and host status is obvious when something is wrong).
  • Remote sessions are stable, responsive, and don’t silently do the wrong thing (model selection, connection, retries).
  • We can actually support it without asking you to become tech support for your own computers.

We’ll post smaller updates as things land (although I expect the next news update to be an update release announcement).

Thanks for sticking around - reply on bluesky link incoming or email contact@ment.tools with anything you want prioritized. If anyone is interested in potentially testing the next build early via TestFlight, let me know-when I have something worth sharing I’ll reach out.