ErmosIQ™. The intelligence layer inside every Ermos unit.

It learns your business, and never leaves the building. Every question your team asks, every correction they make, every document they work with: your Ermos unit learns from it all, on hardware behind your own walls.

An AI that gets better at your work, not everyone’s.

Cloud AI models are trained on the internet. They know a little about everything and nothing about your business. ErmosIQ is the learning engine inside every Ermos unit. From the day it’s installed, it begins adapting to the way your business actually works:

Your language

Matter names, client entities, internal shorthand, practice-area terminology.

Your documents

The templates, precedents, and file structures your team uses every day.

Your preferences

How your people phrase things, what a “good answer” looks like in your business, which sources they trust.

Your corrections

When a user refines or overrides an output, ErmosIQ weights that feedback into future responses.

Diagram: every interaction is weighted, with accepted answers reinforced and rejected ones down-weighted

Six months in, your Ermos unit behaves like a well-briefed member of your team.

The longer it runs, the wider the gap.

Chart: capability rising from week one through month six and beyond as ErmosIQ compounds
Week 1

Your Ermos unit is installed with your document base indexed. Useful from day one: search, drafting, summarisation across your own files.

Month 1

Terminology and entities locked in. The system recognises your matters, clients, and internal language. Fewer clarifying questions, faster answers.

Month 3

Preference weighting is visible. Drafts come back closer to house style. Users edit less and accept more.

Month 6+

Adaptation layers reflect your practice areas. The unit handles your team’s recurring work patterns with a fluency no off-the-shelf model can match, because no off-the-shelf model has seen your work.

Every month an Ermos unit runs, the asset gets more valuable, and everything it builds up belongs to your business.

Four layers of on-device adaptation.

Diagram of the four adaptation layers: knowledge indexing, terminology, preference weighting and model adaptation
1

Knowledge indexing

Retrieval gets sharper the longer the box runs.

ErmosIQ builds and continuously refines a private index of your documents and data. New files are incorporated automatically; stale material is down-weighted.

2

Terminology & entities

It stops guessing and starts recognising.

The system learns the names, acronyms, and phrasing unique to your business and your clients.

3

Preference weighting

Every interaction is a signal.

Accepted answers are reinforced. Edited or rejected answers are down-weighted. Over time, output style converges on your own standards without anyone writing a style guide.

4

Model adaptation

Tuned to your domain, on the unit itself.

Periodically, ErmosIQ consolidates what it has learned into lightweight adaptation layers applied to the underlying model, typically overnight, with no external connection required.

The learning never leaves. Neither does the data.

This is the part cloud providers can’t offer. When a cloud AI “learns from your usage”, your usage is on their servers. Your prompts, your documents and your corrections feed a model shared with the world. With ErmosIQ:

The sovereignty guarantee: all learning happens on the unit, with the model, index and adaptation layers on hardware inside your premises. Nothing is transmitted: the unit has no egress, with no telemetry, no usage analytics and no training data leaving your network. Your improvements belong to you: the intelligence your business builds through use is your competitive asset, not a contribution to someone else's model.

Your competitors using cloud AI are training their provider’s model.

Your business is training its own.

Learning without the compliance headache.

For legal, accounting, medical and financial professionals, “the AI learns from your data” is usually a red flag. ErmosIQ is designed so it’s a feature you can put in front of your compliance officer:

No third-party data processing

Client data is never disclosed to an AI vendor, because there is no vendor in the loop at runtime.

Auditable boundary

The learning system’s entire footprint is one physical unit. Your security team can point to it.

Client confidentiality preserved

Privileged and confidential material informs your private model only. It cannot surface in anyone else’s.

Managed by Ermos, owned in outcome by you

We maintain the unit and the learning system under your service agreement. The knowledge it accumulates serves your business alone.

Straight answers about the learning.

Does ErmosIQ send anything back to Ermos?
No. The unit operates with no egress. Learning is performed entirely on the box. Ermos technicians service the unit on-site under your supervision and your service agreement.
Can it learn something it shouldn't repeat?
ErmosIQ learns within your own boundary: everything it learns came from your data and stays behind your access controls. Role-based access on the unit governs who can see what, just as with your document management system.
What happens to the learning at end of term?
Under your agreement, units returned to Ermos are cryptographically wiped before redeployment. Your adaptation layers and indexes are destroyed with the wipe. Nothing your team taught the system survives to another client.
Does learning slow the system down?
No. Consolidation runs are scheduled outside working hours. During the day, the unit serves your team at full speed.
Is this the same as the AI "training on our data" that we've told staff to avoid?
It's the opposite of the thing your policy is guarding against. Those policies exist because public AI tools absorb your data into a shared model. ErmosIQ keeps the learning inside a machine you control, benefiting only your business.

Every business gets the same cloud model. Yours gets its own.

ErmosIQ is included with every Ermos unit: no add-on fee, no tiers. It starts learning the day the unit is installed.