The engine · built by Whater.org, a foundation in formation

Local Layer for human-first AI.

A reflective control layer for AI systems. Multiple top-tier models work on your behalf. Your records stay yours.

Privacy is architecture-based, not policy-based.

LocalLayer is a working framework, not a product launch. The mailing list is for builders, investors, and partners who want to see it evolve.

LocalLayer is the sixth layer of AI.

The conventional AI stack has five layers. We add a sixth: a local, reflective, user-governed overlay.

Layer	Description	Examples
1. Infrastructure	Cloud and compute	AWS, Azure, GCP
2. Data	Training datasets	Proprietary corpora, medical literature
3. Model	ML and LLMs	GPT, Claude, Gemini
4. Inference	Real-time reasoning	APIs, local engines
5. Policy	Governance, guardrails	GDPR, HIPAA, EU AI Act
6. EveR Local Layer	Local, reflective, user-governed overlay	AI Doctor Ben

The EveR Local Layer is a sixth-tier AI architecture that enables reflective, local, and user-governed intelligence, bringing privacy, transparency, and ethical control directly to the individual.

What changes when you add a sixth layer.

Capability	Conventional AI	EveR Local Layer
Data Storage	Cloud-based	Local-first
Governance	Provider-controlled	User-governed
Ethics	Static	Selectable and versioned
Transparency	Limited	Auditable
Personalization	Centralized	Reflective and local
Privacy	Policy-based	Architecture-based
Compliance	Reactive	Designed in
Data Ownership	Platform	User-centric

The "Ethics: selectable and versioned" row is the live link to our second paper, Governance on Demand (GoD).

One reflective control room. Whatever data you bring in. Whatever AI we use.

LocalLayer sits between you and the AI services you use. It decides what can leave your device, what gets redacted first, which model to trust for which question, and it logs every answer so the system can improve without ever needing to see your raw records.

The architecture is published in our paper The EveR Local Layer, a reflective, local self-training layer for AI assistants. A second paper, Governance on Demand (GoD), sets out the dynamic ethical frameworks the layer enforces. The concept is open. The implementation details, routing rules, system prompts, behavioural principles, are proprietary, on purpose.

Read the paper →

Outer ring: All for your output, the experience the member feels.

Middle ring: Tools and engines, RedactUS, multi-LLM routing, storage, multi-modal prompts, decision tools.

Centre: Local Layer, reflective governance, EveR self-training, audit logs, Community Ethics Modules.

The five jobs of the Local Layer.

Job	What it does
Redact	Strips personal identifiers before any prompt leaves the device. Powered by RedactUS, our open-source engine.
Route	Picks the right model for the right question. Multiple top-tier LLMs work on your behalf, never one black box.
Reflect	Pre-input and post-output checkpoints. The system asks itself whether the answer matched your context, your history, and your stated intent.
Record	Versioned, local audit logs. Every answer is improvable without your raw records ever leaving you.
Govern	Community Ethics Modules. User-selectable, versioned, auditable ethical frameworks. This is the live implementation of Governance on Demand (GoD).

Concept-level only. We publish the what and the why. The how is proprietary.

Working today, in public

The tools that prove LocalLayer is real.

RedactUS

The redaction engine that runs entirely in your browser. Strip personal identifiers from any text before it reaches ChatGPT, Claude, Perplexity, or Gemini. One HTML file. No installation. No internet required. Works offline forever. MIT licensed.

20+ pattern types across UK, US, and EU regions. Names, addresses, NHS numbers, SSNs, MRNs, insurance IDs, clinician and hospital names. The standalone version is downloadable today. The browser extension and visual "what AI sees" preview are next.

See RedactUS on GitHub →

Rocketbot

Find why your content is failing AI visibility, and fix it. An answer-engine optimisation tool that writes and updates content for the way AI systems retrieve and answer questions. Free wedge for the AEO/SEO audience, with a paid Pro tier to follow. In development.

See the repo →

AI Doctor Ben

The first paid product on LocalLayer. See the featured section below.

Jump to the product →

If you want to see how the engine thinks, RedactUS is the door. If you want to see how the engine ships, AI Doctor Ben is the door.

LocalLayer in production

AI Doctor Ben runs on LocalLayer.

Feature	What it does
Local-First Data Storage	Health records stored on-device (IndexedDB)
Privacy Guard	Built-in PII redaction filter
Audit Logging	Timestamped governance records
Consent Management	Explicit GDPR-aligned permissions
Reflective AI Behaviour	Personalised local adaptation
User-Controlled Data Lifecycle	Clear data ownership and deletion controls

Multiple AI perspectives, one reflective layer, records stay on the device.

The "EveR LL Governed" badge in the screenshot is not marketing, it's the live audit point.

The first paid product built on this architecture.

Try AI Doctor Ben →

AI Doctor Ben upload card, EveR LL Governed, device-isolated storage, records stay on your device

Who LocalLayer is for.

For builders

Building a privacy-first AI product in another domain, finance, education, legal, family records? LocalLayer is the framework we wish we'd had. Follow the build; later, we'll talk licensing.

For investors

One engine, multiple products. AI Doctor Ben is the first paid implementation. Rocketbot is the next wedge product, targeting the AEO/SEO market. The architecture is documented in two papers (EveR LL and GoD) with a third public-facing paper in development. The IP boundary is clear.

For press and researchers

The research stream is three papers. Paper 1, EveR Local Layer, is the engineering anchor. Paper 2, Governance on Demand (GoD), is the ethics argument. Paper 3, Why AI Needs Layers Between You and AI, is the public-facing case.

Built by Whater.org

I'm Ben Bacon. LocalLayer started as a working diagram on the back of an AI Doctor Ben architecture review and turned into a published paper because the same question kept coming up: how do you use multiple top-tier AI models for a person without sending their data anywhere?

The answer is a reflective local layer. It's deceptively simple as a diagram and quite a lot of work in practice. AI Doctor Ben is the first product built on it. RedactUS is the first piece of it released to the world. Rocketbot is the second. The research follows the same shape: three papers, EveR LL, GoD, and Why AI Needs Layers Between You and AI. The rest is being built in public, slowly and carefully.

If that's the kind of build you want to watch, the list is the right place.

About the founder on LinkedIn →