Your questions, answered.

For free tools and resources, try RKC for code and text documentation, Sinter’s community preview for local research and handovers, or SubjectNest for lessons and activities.

Ask us to set up a workflow, train you to maintain it or check whether AI suits the task with the 10-Day AI Fit Test. We also assess deployment projects, technical evidence reviews, supported Fracture conversions and applied research. We agree the work and fee before starting.

You can ask to see the Discord demonstration. Self-service installation and paid activation are still in development.

Tell us what keeps going wrong, the tools you use and what you need to do once it’s sorted. A simple, non-sensitive example is enough. We’ll agree the work, price, checks and timing before starting.

Small-business and Sinter setup is A$110/hour including GST, charged for actual minutes with no one-hour minimum. We agree the work and a maximum fee in writing before starting. Ongoing care is optional. RKC and Sinter remain free.

The standard 10-Day AI Fit Test is A$5,280 including GST (A$4,800 ex GST) and is paid before work begins. A$2,400 of the ex-GST fee is credited toward a separately agreed private technical-document setup if both parties sign within 30 calendar days after delivery of the final System Fit Report.

It tests a specific workflow against the target hardware, data requirements and operating conditions. You receive working outputs, the test results and a written recommendation about deployment, changes or further work.

NeuroForge is a research company developing products and delivering selected technical projects. The Discord bot is the leading recurring-product effort. Fracture work currently starts with a supported model and an agreed conversion scope. RKC is available as open-source software.

We welcome proposals for conversion, commercial and funded research pilots. Each project starts with an agreed task, access rights, budget, delivery terms and acceptance criteria.

We choose the model and runtime that fit the task and operating requirements. That can include supported ERAIS components and third-party open-weight models. The proposal identifies the system to be used and how it will be evaluated.

ERAIS Gateway, Training Workbench and Evidence Control are projects still in development. You can see what each does, what needs testing and where a design partner could help finish it.

We want trained AI capability to become reusable: something a team can carry into a new system, combine with other capabilities and keep improving.

ERAIS provides the adaptive architecture and Fracture brings in supported trained components. The commercial aim is to retain the value of training and make development and operating costs more manageable.

ERAIS stands for Efficient Routed Adaptive Inference System.

Think of a system with a growing repertoire of specialists. A task is routed to the relevant components, which use shared context and return their results.

A new capability can be trained or imported into that repertoire. We then test how it works with the existing system and measure the quality and computation involved.

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.

Recorded work includes Qwen feed-forward reconstruction, speech components and supported game-control policies. A proposed customer conversion starts with model compatibility and a comparison of the behaviour, state and resource use that matter to the task.

The retained long-run record contains two native-byte training arms, each completing 262,400 updates and 536.3M valid byte-level targets on four CPU threads.

The richer cognitive stack improved on changing training-time checks. The fixed evaluation did not establish a quality advantage. That result guides the next transfer and comparison experiments; the record retains its original date and conditions.

Real tasks involve speech, images, context and action. The architecture needs ways to connect those inputs and select the relevant computation.

A recorded integration test processed real audio and labelled video while preserving the frozen language core and inactive capability exactly. Broader multimodal quality is evaluated for each model and workload.

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.

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.

Lloyd uses AI across research, implementation, testing and analysis. He sets the direction, designs the experiments and decides what the results justify. The intended team combines specialist human expertise with the same high-context use of AI tools.

Reusable conversion code, knowledge of model behaviour, evaluation tools and a record of successful and failed approaches. The aim is to make each new implementation more efficient and turn that technical base into products customers value.

NeuroForge is open to aligned funding conversations, but no fixed raise amount is currently published. Near-term priorities are qualified paid delivery, evidence-backed research and product milestones. Funding discussions do not imply committed capital, a grant, a partnership or an offer of securities.

NeuroForge is founder-led and pre-revenue, developing its first external commercial engagements. The published company record identifies completed relationships and funding separately from proposals and discussions.

ERAIS is an active research and product-development programme. Public experiments and previews expose specific capabilities and report current availability. A production deployment requires workload-specific quality, reliability, security and operating tests, with service terms agreed for that deployment.

Each record states the model, inputs, method and result. The current engineering summary covers running source models, exact-state checks and component CPU measurements. Earlier conversion and training experiments retain their dates, datasets and limitations.

Yes. A well-run experiment can show that a proposed change is slower, less accurate or unsuitable for the task. That answer can save a team from a much larger unsuccessful investment. We keep it in the record.

NeuroForge provides technical evaluation and engineering recommendations. Legal, clinical, safety and regulatory certification require the relevant qualified professional or authority.

Use a non-sensitive description or public example. Keep credentials, patient information, confidential customer records and controlled material out of the first message. We agree access rights and data handling before receiving project material.

The Doom experiments include complete native neuronal execution and an adaptive executive. Minecraft has a separate perception-and-action stack. Unified World and the NeuroForge model API use the shared World owner, and Story Lab uses a natively trained language checkpoint. Each experience reports its own availability and model scope.

Start with Current results for the latest engineering record, then follow a result into its source data. The evidence library contains earlier experiments. The claim register retains reviewed statements and their dates; machine resources describe the public interfaces.