Site index
Find your way around NeuroForge.
Jump to a tool or service, explore the experiments, or look up the research.
Tools & services
- RKCFind your way through code and documents on your own computer, with links back to the original files.
- SinterOrganise meeting notes, funding research and handovers on your device. Export a backup for the next volunteer.
- Bots and game servicesExplore the Discord conversation pilot and game AI work, with current product boundaries.
- SubjectNestFind free lessons, printables and home play by year or subject. Download resources to adapt or explore the learning prototype.
- Small businessGet help tracking customer enquiries, refreshing reports or finding answers in your business documents. Setup and training for one agreed task.
- ConsultingFind out whether AI helps with your job, whether a release is ready or how your team can run it privately.
- EnterpriseGet help with a costly or slow AI setup. Review an existing model, run a task privately or test a research question.
ERAIS
- Meet ERAISThe adaptive architecture: specialist capability, shared state and selective computation.
- Fracture conversionConvert supported learned capability into ERAIS, validate it and build toward modular upgrades.
- Inside the architectureThe deeper explanation: native training, Fracture and shared cognition.
- API guideConnect text generation to your app. Discover the selected model, request shapes and current limits.
- Research programmeThe experiments, results and open questions driving ERAIS and Fracture.
Experiments
- Unified WorldOpen a session and see which inputs are available.
- Minecraft AIWatch an experimental AI player in Minecraft, with its current activity, limitations and development status.
- Doom AICompare three experimental players in Doom and inspect their current activity and model differences.
- Fly LabExplore the simulated fly body, brain activity and vision, with live measurements and source details.
- Story LabTry a short story with a small experimental language model when the shared service is available.
- Reasoning LabExplore symbolic reasoning examples and inspect the steps behind an answer.
- Speech comparisonsListen to recorded speech comparisons and read their methods and current limitations.
Evidence
- Current resultsCurrent source models, engineering measurements and the records behind them.
- Evidence libraryMeasurements, source versions, methods and limitations from the research archive.
- Public claim registerTechnical statements with their review dates, source records and scope.
- Model fractureThe bounded model-conversion evidence trail and current limits.
Company
- NeuroForgeMeet the lab, find its tools and services, and see what the research is working towards.
- FounderHis background, approach and what he is building.
- Funding & partnersSee the current results, funding priorities and ways to get involved.
- MethodHow we compare systems, record failures and decide what to develop next.
- Help & answersAnswers about NeuroForge, ERAIS, Fracture, products, readiness, pricing, privacy, and negative results.
- ContactProduct setup, business workflows, technical work, research or media.
Complete evidence index
- Engineering inventory snapshotA timestamped repository-scale, evidence-inventory, and nine-claim presentation snapshot retained for historical traceability.
- Review readinessProof inventory, public claim posture, release checklist, and risk-register status.
- Quality gatesConfiguration validation, critical gate coverage, phase-control checks, and API observability posture.
- Evidence curvesPublic aggregate memory and smoke-loss guardrail curves across the approved experimental scales.
- Exact sparse-plan constructionExact plan-identity checks and host-local construction timing for one sparse-attention planning component.
- Extreme-context decode ladderA replicated current-host context ladder across three CPU-only synthetic candidates, with cache-representation counts, semantic diagnostics, and admission boundaries.
- Model fracture milestoneReal Qwen3 0.6B fracture, exact K32 engineering reconstruction, and a completed protected four-way trace for material-sparsity alignment.
- Terminal evidence handoffAn exact-source, sanitized public bundle joining current phase and claim reconciliation with a 2M synthetic context candidate and two retained sparse-geometry negatives.
- Training probeStep counts, throughput ranges, parameter footprint, and pair-work reduction from bounded probes.
- Continued trainingLoss signals, stability controls, generation gates, reload behavior, and token-scale receipts.
- Full-stack trainingBounded throughput, loss, safety-counter, warning, and claim-boundary evidence from the full training path.
- 100M scale matrixActive parameters, evaluation loss, memory peak, and throughput-ratio results at the approved scale point.
- Resource outlookProjected active share, stored parameters, decode bandwidth, optimizer state, and weight-stream requirements.
- Diligence readinessArtifact coverage, claim posture, clean-room controls, security review, and controlled gates.