Avery NXR is local-first

If you've spent any time in the last decade shipping internal tools, you've probably ended up with the same shape of problem. A spreadsheet grew into a team-wide thing. The team-wide thing needed a real database. The real database needed an admin UI. The admin UI needed authentication, then audit logs, then a notifications channel, then a webhook, then an export job. Six months later you have a Next.js app that nobody on the team wanted to write but everyone depends on.

Avery NXR exists to compress that loop. You describe what you want; the local Small Language Model plans the work; versioned generator plugins write the code; the result is a real Next.js 15 app with a real Prisma schema and a real Tailwind UI you can deploy anywhere. No SaaS lock-in. No data egress.

What "local-first" means here

Three things, concretely:

1. The compute that builds your app runs on your machine. When you ask Avery NXR to scaffold an order-tracking app, the SLM, the schema composer, and the generator plugins all execute locally. We don't see your prompts, your schema, or your generated code.
2. The app you build runs anywhere — usually not on us. The output is a plain Next.js project. Push it to Vercel, Railway, your on-prem Kubernetes, a Raspberry Pi — same source, same Prisma schema, same migrations.
3. The cloud is opt-in. We have a central server that handles billing, licensing, and a curated SLM catalog. It does not see your project source unless you explicitly enable Consult Mode for a specific request, and even then the traffic goes direct from your desktop to the cloud LLM provider you choose — never through us.

What you lose, and what you gain

Local-first isn't free. The tradeoffs are real:

• You need a machine that can run a 7B-parameter SLM. Apple Silicon and modern PCs handle it. A 2018 Chromebook does not.
• Local inference is slower than the fastest cloud frontier model. Often meaningfully so for very long prompts. Consult Mode exists for the cases where this matters.
• Updates land via a signed auto-updater rather than a transparent cloud rollout. Most days that's a feature; occasionally it's a wait.

What you get:

• Zero data egress by default. Your code, your prompts, and your generated apps stay on disk.
• Full control over deployment. No "you have to deploy through us" vendor pattern. Generated apps are yours, fully.
• Predictable costs. No per-request billing on the local model. Pay for Pro / Enterprise tiers if you want Consult Mode + Plugin Forge + workflow bundles; otherwise the local SLM ships free in every tier.

What's next

The roadmap from here is mostly about closing the gap between "the app generator works" and "the app generator handles your specific business" without you having to learn its internals. More archetypes. More provider integrations. Better failure modes. We'll write about each capability as it matures — every post on this blog is a walkthrough of something you can actually do with Avery NXR after you install it.

Welcome to local-first software development. We're glad you're here.