Animated illustrationFOUNDATIONS · Luna walks the room while a rack smokes.
Foundations

The manifesto states the why. This document states the how.

What HME84 is in concrete terms, how the process works, the framework we use, and what we honestly do not do.

A model can go offline. The work shouldn’t have to start over.

HME84 orchestration: judgment, system and execution across multiple engines. Models change; judgment remains.

When an engine goes down

  1. Engine unavailableFirst the failure is recorded and the state is saved, with its partial results.
  2. Context keptThe goal, the decisions, the sources and the point of work are kept, if they were recorded.
  3. Validated handover or a safe pauseWe check whether there is an authorised alternative with enough context and capability. If there is no safe handover, the work pauses and the block is made visible.

01

Models

Explore, compare, reason. From different countries and providers.

USA

  • ChatGPT
  • Claude
  • Gemini
  • Gemma
  • Perplexity

Europe

  • Mistral

China

  • Qwen
  • DeepSeek
  • GLM

Judgment and orchestration

HME84

HME84 orchestrator

Model is not Agent is not System

The model provides capability. The agent takes responsibility. The system coordinates the whole.

02

Building and output

Turn judgment into execution.

Code

  • Claude Code
  • Codex
  • Cursor
  • Antigravity

Tools

  • Canva
  • CodeH5 (built at HME84)

Agents

  • OpenClaw
  • Hermes
  • Agents84 (built at HME84)

Built at HME84.

03

Our own runtime

Our own foundation to reduce dependence on external services.

  1. Local modelsGemma · Qwen · MistralThey run on our own machine.
  2. KnowledgeXerebro84Our own index, with provenance and no network.
  3. CoordinationMv2The mailbox agents use to hand work to each other.
  4. Memory and evidenceRecordsWhat was decided, what was checked and what is missing.
  • Local first
  • Cloud only with permission
  • Resilience by design
  • Deep integration
A conceptual view, updated on 12 October 2026. The applications are examples of use depending on the task, not an inventory of active integrations; their names belong to their owners. It does not certify availability or automatic handover: continuity depends on the context kept and the capability validated.

When an engine goes down

What happens to the work?

01 The model changes

The goal, the decisions, the sources and the point of work are kept, if they were recorded. What the new engine can do, and with which permissions, is checked again.

The handover is validated before continuing.

02 An engine stops being available

First the failure is recorded and the state is saved, with its partial results. Then we check whether there is an authorised alternative with enough context and capability.

If there is no safe handover, the work pauses and the block is made visible.

03 A task looks finished

Delivering a file is one step. A passing test does not by itself prove that it works in production or that the person has accepted it.

Done means meeting the result and checking it.

For your project: documented decisions, transferable deliverables and less dependence on any single tool.

A studio that steps in before building.

HME84 does not start by selling a solution. It starts by listening to a real problem — and deciding whether technology has something useful to contribute, or whether the honest answer is to leave things alone for now.

We combine technical judgment, editorial judgment and brand sensitivity. They are not separate layers: they are the same perspective applied to the same problem.

The value is not in the code. It is in the reading that precedes the code.

HME84 works with businesses that already have reality: customers, operations, friction, pending decisions. A bakery, a pet store, a clinic, or a boutique hotel do not need the same scope — but they do need the same judgment.

Loop84 / Root84 / Glow84.

  • Loop84 · Act, correct, do not react.
  • Root84 · Strong roots before growth.
  • Glow84 · Shine without disguise.
What changes in each phase

Direction before reaction. Roots before growth. Shine without disguise.

Three phases, one way of judging. Not every project reaches the third. Sometimes Loop84 is enough — and the most honest deliverable is clarity without spending on what you do not need.

Loop84 tells you how to act, what to correct, and when not to react. It turns noise, urgency, and technology anxiety into a clear next action.

We listen without an agenda. We map what actually exists: what is used, what is paid for, what repeats, what creates friction. We separate real urgency from digital fashion. The result is not always "something needs to be built" — sometimes the most honest diagnosis is clarity with no additional spend.

Assessment comes before building: the other steps depend on what we find here. The only exception is GBC Hotfix, a one-off, tightly scoped intervention on a specific problem; if the problem turns out to be bigger, we document it and propose the right path.

Root84 organizes what sustains the business: tools, processes, subscriptions, costs, messaging, flows, and operational decisions. Before adding new layers, what already exists has to work well.

Root84 does not build anything new. It makes what already exists work better — and that often matters more than any additional tool. An organized stack is easier to run, to pay for and to understand.

This phase focuses on consolidating what is already there, not on adding tools.

Glow84 builds the visible layer of the business without disguising it. Digital presence, microsites, assistants, flows, bots, and tools that express what you already are better — not what you think you should look like.

When the diagnosis confirms that building makes sense, we build. Lightweight does not mean lesser: it means transferable. The client can understand it, maintain it, and not get trapped in blind dependency. When a project uses AI, we define the task it supports, the information it receives, how its output is reviewed and when a person steps in.

Geek enough to build it. Chic enough to make it belong.

What working with HME84 looks like.

  • First conversation
  • Diagnosis and map
  • Clear deliverable
  • Real next step
The full process

You do not need to arrive with a perfect brief or a finished idea. The first conversation is to understand the real problem — not to sell a service. If HME84 is not what you need, we say so on that same call.

We listen, ask questions, review what exists. We separate what is urgent from what is important, and what is real from what is fashionable noise. The result is an honest reading of the case — not a pitch to convince you to build something.

There is always a concrete deliverable: a blueprint, a stack review, a suggested path. The document is yours. You can take it to another team, execute it yourself, or use it to make an informed decision.

If it makes sense to keep working together, we define the next step with clear scope and expectations. If it does not, we say that too. The proposal separates the development price from running costs: third-party APIs, domains, licenses and subscriptions are quoted separately. We work remotely, by video call, with direct communication and no intermediaries.

What HME84 does not do.
  • Generic consultancy without real technical judgment.
  • Inflated digital transformation to justify budget.
  • Custom development as the default answer to any problem.
  • Empty packages designed to look complete.
  • Unnecessary complexity when a simple tool solves the case.
  • “AI solution” promises without a real prior assessment.
  • Dependency on proprietary tools that trap the client.
  • Projects where the deliverable cannot be used without HME84.
View services →