Straight answers. No inflated claims.

Modern AI often gets more capable by making more machinery part of every task. NeuroForge is building around a different rule: use the right tools and the right amount of effort for the task, then leave the rest alone.

The long-term goal is intelligence that can keep expanding without making every thought proportionally more expensive.

ERAIS stands for Efficient Routed Adaptive Inference System. This is the only current approved expansion used by NeuroForge.

Imagine you want to add one house to a neighbourhood. The brute-force answer would knock down the whole neighbourhood and rebuild it with the new house included.

ERAIS is the other approach: build the new house, connect it and leave the rest standing. Add what is missing. Use it when it is needed.

ERAIS is an expandable cognitive architecture. Language modelling is one capability. Implemented paths also process vision, audio, memory and action outcomes, with adaptive scheduling and persistent state. The capabilities available to a visitor depend on the particular model and interface.

Fracture converts learned model components into modular ERAIS expert structures, building on capability already learned by conventional AI models.

Implemented conversions include learned speech projections and complete native checkpoint replay for supported Minecraft model families. Persistent router learning and full-compute reference agreement have also been exercised. Each result has its own model and evaluation boundary; useful quality, efficiency and deployment readiness must be established for the intended workload.

Two long language-training runs completed 262,400 updates and 536.3M real byte-level training targets each on four CPU threads.

The richer ERAIS system looked better throughout the changing checks, but that advantage did not hold on the fixed final test. We did not claim a quality win.

Important work is spoken, seen and done in motion. ERAIS is intended to treat modalities as selectable capability paths rather than forcing every task through one interface.

Real audio and labelled video have traversed the shared sparse system while the frozen language core and inactive capability stayed exact in the declared test. That is a modularity signal, not a production multimodal-quality claim.

ERAIS training, Fracture conversion and multimodal execution have been demonstrated. The next evaluations measure broader quality, efficiency at matched quality, scale, independent reproducibility and readiness for specific deployments. Each published result links to the experiment that supports it.

Small-business services help you keep running enquiry, reporting, knowledge and AI workflows in your own tools. Actual customer-specific setup and training is A$110/hour including GST (A$100 ex GST), with a written maximum after a relevant demonstration. Ordinary updates need no subscription; developing workflows are labelled on the service page.

For a larger AI deployment decision, the 10-Day AI Fit Test is A$5,280 including GST (A$4,800 ex GST). It tests one real workflow and leaves a written deploy, change, remediate, rescope or stop decision.

Start with one everyday task and your current platform. For a small-business service, bring a public or fictional example to a free 15-minute fit call. We confirm the job, price, acceptance checks and delivery date before payment. A larger deployment decision may suit the 10-Day AI Fit Test.

The standard 10-Day AI Fit Test is A$5,280 including GST (A$4,800 ex GST) and is paid before work begins. If the separately scoped NF-A1 deployment stage is signed within 30 calendar days, A$2,400 of the Fit Test fee is credited to it.

Access to models does not answer which combination fits this workflow, data boundary, machine, operating environment and owner. The Fit Test buys an integrated deploy-or-don’t decision with failures left visible—not allegiance to one model or platform.

Small-business setups and the Fit Test are founder-delivered services with fixed scopes and working handovers. Deeper paths are separately scoped. RKC is released open-source software; its optional setup service pays for configuration and training. ERAIS remains a research and product programme whose deployment claims depend on the measured result.

Qualification is open for the first bounded commercial, design-partner and funded research pilots. Discovery or a proposal is not an active pilot. Work starts only after scope, rights, funding, acceptance criteria, operators and an agreement close.

No. Commercial work uses the qualified model and runtime that best fit the approved workflow and boundary, including third-party open-weight components. Native ERAIS is used only where its current capability and evidence justify it.

ERAIS Gateway, Training Workbench and Evidence Control are gated candidate surfaces. Discovery and non-binding design work may occur, but interest does not make them production offerings or existing partnerships.

NeuroForge is the brand. NeuroforgeIO Pty Ltd is the Australian company, ACN 701 314 329 and ABN 87 701 314 329. It is incorporated, founder-led, pre-revenue and based in Queensland.

Lloyd Handyside is the full-time founder. His multidisciplinary path spans biotechnology and medical research exposure, field and industrial chemical AI, big data, programming and the ERAIS architecture.

AI amplifies expertise; it does not remove human direction or accountability. Lloyd uses AI to expand context, implementation, analysis and iteration while retaining judgement, synthesis, experiment design and stop decisions.

Every useful result improves the next experiment. Every failed result removes a dead end. ERAIS, Fracture, the Observatory, the tools and the AI-amplified way of working learn together. A follower would have to catch a moving system—not copy one codebase.

Yes. NeuroforgeIO Pty Ltd is seeking A$695,000 for an 18-month proof and commercial-validation programme: independent technical conclusions, first paid Fit Tests, customer-operated attempts, delivery capacity and a clear next-capital or stop branch. This public material is not a securities offer.

The company currently claims zero customers, zero paid pilots, A$0 revenue and A$0 external capital. Pilot qualification, an application and early research discussions are not represented as customers, partnerships, awards, funding or independent validation.

Public previews run today, including Unified World, the Fracture API and Story Lab. No public production-readiness claim is approved: broad device support, production service levels and commercial native-model readiness still require their model, security, operations and evidence gates to pass.

Each evidence pack supports only the result it actually measured. A small systems test is not automatically a quality comparison. A working mechanism is not automatically a product. An internal record is not automatically independent validation.

Yes. Do not deploy, hold release, the mechanism failed, or the hypothesis did not clear the baseline can all be successful outcomes when the protocol ran correctly and the result resolves the agreed decision.

No. NeuroForge does not currently provide legal, privacy, clinical, safety, cyber, engineering or regulatory certification. An evidence audit is a technical reconstruction and release recommendation, not accreditation or legal assurance.

Do not include classified, patient-identifiable, export-controlled, safety-critical, critical-infrastructure, credential or other protected material. A project proceeds only after lawful rights and an explicit handling boundary exist.

Unified World and the public Fracture API use the resident service, combining native ERAIS components with retained donor components. Story Lab uses a separate model trained natively. The older chat workspace remains held; its development fixtures do not establish ERAIS capability. Public previews are distinct from a production-readiness or service-level guarantee.

Use the claim register for approved wording, Progress and proof for the dated executive state, the evidence library for methods and limits, and the machine resources for current public contracts. The reviewed public-results snapshot is deliberately separate from the moving ERAIS repository head.