Etheraio

Revenue share and crypto mining

Run your installed Etheraio AI pipeline for game designs, images, 3D prototypes and checked learning. Scientific tools verify research separately. Hardware-sized chunks respect your resource choices.

MINING OFFNo time limit

Ready when you are.

Preparing the native engine…

The combined Mine uses spare CPU and GPU within the limits you set for AI, research, game production and training. It can also run checked crypto mining alongside that work when you opt in.

Play the selected Unreal project as it develops. Stop mining first. This PC needs the existing engine and game files; public game downloads are not yet available.

Supported asset checks run automatically. Local AI records visible defects and repair suggestions.

Auto checks every compute stage against the detected CPU, GPU and available AI accelerators. It uses a matching checked runtime where available—currently AMD ROCm or NVIDIA CUDA for specialist model work—and otherwise follows your CPU-fallback choice. A detected NPU/AI chip is not treated as usable until its runtime and stage pass. Hardware-sized batches and the selected effort limit avoid reserving unavailable memory; Maximum requests all permitted effort but does not force a constant 100% or override heat, OS or user limits.

Automatic experiment review waits for technical checks and a source-bound local AI review. Acceptance admits that asset version to the experiment; public release and AAA quality need further evidence.

Continues when minimized. Closing the app stops mining. Nothing starts automatically.

VERIFIED CONTRIBUTIONSRevenue shareYour verified compute ÷ the network’s verified compute at the funded withdrawal snapshot.
Verified jobs · account total—
Verified jobs this session0 jobs · 0 reward blocks
Local processing—
Next available job—

MAKE ROOM FOR YOUR DAY

Choose your pace

Normal
25 / 100
OffMaximum

Detecting the CPU…

Off
OffMaximum

GPU processing uses fixed, verified kernels.

Checks each GPU and combined processing without mining or submitting reward claims.

AI, research and game work take priority. Device-sized batches, thread limits and pacing follow your shared CPU/GPU targets. Checked crypto uses measured spare capacity and yields when AI needs it. Quality, model and verification settings remain fixed. Whole-device readings include other apps; uninterruptible kernels can briefly exceed a utilisation target. Stop ends the owned work.

Choose CPU and GPU targets from 0 to 100%. Auto compares checked same-model text execution, including reload time, and uses compatible media stages with memory-sized batches or offloading. Model, context, image size, 3D detail and verification stay fixed. If the selected quality cannot fit, the job waits. More utilisation alone does not improve quality, and instantaneous activity can exceed a pacing target.

Requested and measured utilisation appears after native telemetry is available.

Keeps the original image model, 1024 × 1024 output and four steps. Requires at least 2 GiB available for the worker plus OS/app memory, and the original 16 GB model files on disk. A PC with only 2 GB total RAM is not verified. Full-image low-memory validation is still experimental.

Save controls to update your mine. An active mine resumes with the new choices.

OPTIONAL LARGE LOCAL MODEL

Etheraio Core 27B

The current Etheraio Core profile with checked weights and device-sized CPU/GPU processing. Saved design parts are assembled with technical checks. Smaller parts reduce task memory; the whole model still needs enough memory. Low-memory phones can use their smaller installed model or an explicitly shared host. Neither perfect assembly nor equal speed on all devices is guaranteed.

The 27B choice waits when memory is insufficient instead of silently substituting a smaller model. The existing low-memory output adapter remains pinned to Qwen3.5. Qwen3.8 uses its own checked inference weights; Qwen3.5 adapters are never reused as Qwen3.8 training. The separate full native LoRA profile still needs its pinned snapshot and memory checks.

Lower-memory PC training reads the existing frozen GGUF model and trains a small output-layer LoRA on the CPU. It does not download a second full-precision model. Checked data, free memory and Start consent are required; this is a scoped experimental adapter, not full-model training.

Low-memory trainer is loading.

Advanced AI runtime & licences

The farm, workspace and permitted shared AI hosts use the same checked request adapters. These profiles do not train or merge weights. Image, 3D, audio and numerical simulation keep their dedicated runtimes.

  • Check local AI models to see installed specialist candidates.

Optional model setup is loading.

Same licensed upstream weights, not a newly trained model. Existing smaller models, selected effort and full-block reward rules remain in use.

ACTUAL DEVICE MEASUREMENTS

Your live compute

CPU activity
Unavailable
GPU activity
Unavailable
Dedicated VRAM in use
Unavailable
Available RAM
Unavailable
Temperature / power
Unavailable

Windows counter sampling starts in the installed app.

GPU activity is the busiest physical engine per adapter across this whole PC, including other apps. Effort controls pace jobs; it does not enforce an exact utilisation percentage.

Sharing is off by default. Device readings are informational and do not earn contribution weight.

ONE START · SHARED CPU + GPU SCHEDULER

Research. Build. Learn.

Choose the foreground priority below. Game-first prioritises new game assets while supported medicine and science work continues on available resources. Existing network verification and rewards stay unchanged.

One choice includes medicine-first network work, all eleven installed medicine models and structure/conformation checks, original assay-model training, public Qwen adapter training with automatic upload and reward evaluation, game assets and rendering, installed local AI, materials learning and calculation audits. Private molecule screening is included only where an existing owner setup is available; its owner sign-in is still required.

Uses your current resource and battery settings. Prepared work uploads to its existing account storage; private chemistry stays encrypted and owner-only. Sharing research or device measurements remains your separate choice. Missing tools, models and scientific checks stay unavailable. This does not configure clinical safety or treatment approval.

Group farming — create or join a node group

Invite people to farm together using their own devices. Each device keeps its own keys and checked work receipts. Joining does not start computing or share private files, chats or chemistry.

Open this panel to load your group.

Etheraio node: contributors 99.9%, Etheraio 0.1%. User-created node: contributors 99.85%, node operator 0.1%, Etheraio 0.05%. Native coins stay separate from cash contribution records; only verified mining shares and real confirmed funds can be paid. Group settlement is not yet connected.

Prepared group policy: 0.1% operator, 0.05% Etheraio, 99.85% contributor before withdrawal fees, for historical work and native crypto separately. Group fee collection is inactive. Current balances and payouts remain unchanged.

Create a group

Join a group

Registered devices, recently seen devices and recent accepted contributors are different counts. A group does not combine GPU memory or turn queue entries into completed work. Invited AI requests have their separate sharing controls.

Medicine checks included automatically

Structure validation, molecular descriptors, PAINS/BRENK/NIH/ZINC alerts and ETKDG/MMFF conformations run with the eleven-model suite: Ames mutation, chromosomal aberrations, liver toxicity, carcinogenicity, developmental toxicity, acute oral toxicity, repeated-dose toxicity, solubility, skin permeation, plasma protein binding and P-gp activity.

These are bounded research predictions with model-domain warnings. Heart-rhythm/hERG, kidney, neurological and immune toxicity, exposure, drug interactions and disease-target efficacy need additional suitable models or evidence. They remain unavailable rather than receiving an invented result.

Choose this once, then Start. Existing work choices are preserved until you change them.

Optional task filters and storage

Knowledge checking starts with eligible public farm work. First-verification weight follows the server policy.

Choose all work types for one combined mine, or select only local AI or game assets. Local-only mode can run without sign-in; it does not claim network rewards. Automatic server checks run with network mining when enabled. Legacy independent audits remain available.

Public medicine model checks run with the same Start/Stop on compatible PCs. They do not need your private vault key.

Original molecule screening trains with public medicine and local training enabled. A held-out report is saved to your private cloud outputs; no clinical approval or reward acceptance.

New public training inputs credit 1,024 weight units only after the uploaded adapter passes the trusted automatic evaluator. 1,048,576 units complete one reward block. Duplicate, failed, private and local-only training earns zero.

Continuous retraining collects fresh checked public evidence while this mine runs.

Checking learned research parameters. Experimental answers still pass through the numerical verifier.

Complete the first checked training cycle, then stop mining to open saved chats, Build and creation previews.

Run a real selected-GPU gradient and separate research-adapter check while the farm is stopped.

Training setup is loading.

Advanced training setup

Use the setup button once to download and check a portable Python runtime and full Qwen3 weights. Automatic selection uses the pinned AMD ROCm 7.2.1 runtime for supported Radeon cards (tested here on RX 9070 XT) or a checked NVIDIA CUDA runtime; other devices can use CPU training. A real GPU gradient check must pass before use. The installer does not contain your account details or private research. Shared Start trains and checks your Etheraio Research candidate alongside medicine-first work when tools, fresh evidence and resources are ready. Later checked public network results feed new training cycles. Batch sizes are measured on the device within your chosen memory budget. Stop and battery/idle settings apply to training too. Repeated data does not cause an endless retraining loop. Every cycle automatically checks source files, finite trained weights and held-out generated answers. A passing numerical-reference benchmark is labelled separately from an experimental candidate; failed checks need no owner review and earn zero. This does not configure clinical or production-art approval.

Game assets target photorealistic production quality. Active generation and render jobs can continue beyond the old time limits while making progress. Stop, device limits and a one-hour inactivity monitor still apply. Elapsed time does not approve an asset or complete a reward block.

Your farm uses the installed local Etheraio AI pipeline. Checking its ready stages…

AI batch checkpoints are saved on this PC. Start resumes unfinished segments and delivery; Stop pauses all work.

Choose your work, then press Start mining below.

Choose this for full CPU/GPU utilisation target on your selected work. Memory budgets adapt to this device; compatible tools and useful assignments are required. Start remains your choice.

Job size follows unique verified work. Shared devices process separate segments; only accepted weight counts toward a full block.

GPU jobs share one queue to avoid VRAM contention. Maximum uses all detected CPU threads when free; the scheduler reserves worker threads while Blender runs.

Assets require installed Blender 5.x; local material AI requires a prepared Etheraio AI runtime. GPU workers run in Turbo, Optimal or Maximum. The original assay model trains once from measured public Tox21 labels, with separate scaffold holdouts. Search-priority learning waits for verified chemistry benchmark outcomes. Negative results lower retry priority only for that subject, model and conditions; they are retained for separately reviewed uses. Continuous Qwen retraining uses the configured offline model and checked public evidence. It does not train private structures or approve medicines or production art. No downloads start automatically.

Blocks from every work sector

Only jobs that meet the active reward acceptance rule credit your account. Reward eligibility is set by the live server policy. Eligible verified calculations, passing network references and unique checked sounds earn the live reward when their policy is enabled. Accepted public AI training contributes 1,024 units per first credited training input; held-out, duplicate and failed inputs earn zero. Points are paid only on a complete 1,048,576-unit block. Unaccepted 3D renders and local-only training still earn zero. They do not inflate your live balance. The asset queue holds up to 16 outputs for review; its storage budget is checked before each job.

No local blocks yet.

EQUAL REWARD BLOCKS

Reading saved block progress…

Checking the active reward policy…

Weights from completed jobs

Sign in to read saved contributions.

Every accepted job is recorded once in your account. These units are already included above. Smaller jobs combine into full reward blocks; leftovers carry forward.

Checking which selected work can currently add weight…

Weights by job
Reading the server’s work weights…

Eligible, unique work accumulates across sectors and devices in your account. Points are paid only when the server records a complete block; any remainder carries forward. Total accepted weight keeps growing even when the progress bar resets after a full block.

Failed checks, repeated inputs, personal chat, private research and unaccepted prototypes have zero reward weight. Local training needs an accepted public-training receipt. Running more of those jobs will not move the reward bar.

The budgets below cover all registered job types. Included stages share their parent budget. Accepted work combines across your Nodes and sectors. Every ordinary reward block has 1,048,576 fixed accounting-weight units; leftovers carry forward. Failed, duplicate or unsupported work adds no weight. Existing balances stay unchanged.

YEARLY REWARD HALVING

Reading live reward…

Next halving: connecting…

Future eligible rewards halve yearly. Existing balances stay unchanged. Failed, incomplete and local reference work does not automatically create funded cash entitlements.

ENTIRE NETWORK

Verified work records: Connecting…

Completed full reward blocks: Reading…

Verified network accounting rate

Accepted work units / second: Connecting…

Verified work units are separate from AI tokens and reward-block weights.

YOUR AI FARM

What your farm is doing

Nothing runs until you press Start.

Research

Supported medicine predictions and public science calculations.

Game content

Original images, 3D models, textures, sounds and supported engine checks.

AI learning

Shared grid work is optional. Join a grid on the website to contribute.

Fresh checked results feed compatible trainers. Candidate improvements require evaluation.

Weekly AI updates are waiting for a checked model.

Work → checks → verified contribution weight. Only eligible new accepted work adds weight. A job, prototype, repair or training run does not automatically create cash earnings. Your saved progress and live rate appear below.

LLM processing: CPU only keeps model layers in RAM; GPU only requires the model to fit permitted GPU memory; combined splits layers across CPU and GPU. Auto compares permitted CPU and planned GPU execution. Small-context builds and long design briefs retain checked parts for retry. The selected model weights still need to fit a compatible local or authorised shared-node runtime.

CASH CONTRIBUTION NETWORK

A separate, verified contribution record.

The cash research-contribution service is at etheraio.com. Use its email and password account to track funded earnings and manage withdrawals in your browser. These native desktop jobs still use the original coordinator; they do not automatically create cash entitlements in the new farm. Existing work records stay preserved.

ETHERAIO WORKSPACE · AI FARM

Your AI output

This PC · last model job

No measured AI output

Generated tokens / second during model generation.

This PC · last 60 seconds

No measured output

Completed output, including idle time.

Network · last 60 seconds

Waiting for reports

Completed output from recent reporting devices.

Actual model output is separate from verified accounting units.

AI batch sizes follow device memory and thread limits, then tune from actual prompt timings. Output quality and reward checks stay unchanged.

Your workspace and farm share the Etheraio inference engine. CPU only, GPU only and combined CPU/GPU use the same selected model. Combined mode splits model layers and can be slower; Auto compares permitted methods. Small-context web builds and long game-design briefs run in checked parts, with completed parts kept for retry. The model weights still need to fit a compatible local or authorised shared-node runtime. Auto can compare permitted CPU/GPU execution.

Optional and off by default. Only aggregate counts are public. Prompts, private chats, files, wallets and chemistry are excluded. This is separate from sharing device details with your private dashboard.

Network totals require opted-in compatible measurements.

Tools & modelsInstall, check and manage your local AI tools.

ONE PIPELINE · PAIRED DEVICES

Take your media studio to your phone.

Etheraio’s 3D, image, motion and face-animation workspaces use the installed desktop capabilities. The phone sends a request and receives its output over a private paired connection. Full model execution stays on the desktop; each stage keeps its existing runtime, licence and memory checks. Motion is currently unconditional and facial retargeting remains experimental.

Your CPU, GPU and memory controls below stay authoritative. Stop farming and close the local workspace before pairing. Shared training never uses these private requests automatically.

Paired processing is off.

OPTIONAL TOOLS · CHECKED DOWNLOADS

Set up your AI specialists

Use Start mining above. It checks compatible tools and offers one download choice for missing packages, then starts ready work with your saved limits. No model paths or separate specialist switches are required.

Advanced package options

One setup prepares checked compatible packages; Start mining then runs ready trainers.

    Package setup and weight training have separate checks. The shared mine trains configured Qwen, original assay and experimental TRELLIS profiles. Other families show the exact missing trainer/data requirements. Voice, transcription and character packages provide inference. Ready concept-art jobs cover vehicles, cities, terrain, characters, interiors, props, spacecraft and materials. AI image references do not replace game-ready meshes, UVs or final art acceptance. No private data is added to shared training.

    One coordinator uses the eligible engines below. Setup keeps your working models and private data preserved. Start runs only the jobs supported by your installed tools, permissions and admitted memory.

    Use the updated Windows app for checked setup. No downloads start automatically.

    Model familySetup and availabilityRequirements

    Your existing Qwen/TRELLIS training setup remains available below. The new media overlay has real CPU voice and transcription checks; Windows AMD speech and transcription use CPU while other GPU work can continue. FLUX needs sufficient RAM and a tested image backend. Kimodo, ARDY, Audio2Face and gated checkpoints keep their own tools, access, licensed-data and acceptance requirements. Preparing source folders or importing a model class does not establish that a model can generate, train, or serve paying users. The planned £4.99 AI subscription stays disabled pending service quality, capacity and billing checks.

    ONE ETHERAIO AI · SPECIALIST ENGINES

    The right specialist for each job

    Etheraio AI pipeline connects the selected mining tools when you press Start.

    Specialists prepare under the shared Start.

    SectorEngineAvailability

    3D setup runs with the shared Start after compatible training tools are ready.

    Optional checked Microsoft geometry/PBR codecs use CPU or a compatible PyTorch GPU. Shared Start trains a separate material head on original controls and decodes its textured mesh. This does not train the full 4B image generator or approve game assets.

    Optional checked generator stages train with shared Start. No download starts automatically.

    Experimental 3D component training joins Start after checked PyTorch setup.

    Your latest completed creation

    A checked concept image appears here when the mine completes one.

    Concept artwork remains a prototype; it does not prove 3D completion or production acceptance. Saved images, audio and transcripts can also be previewed in My saved outputs.

    Your personal game AI

    The local game designer prepares a persistent brief before image and 3D generation.

    Checking accepted asset weight and the current server reward policy.

    Checked public evidence feeds automatic retraining; unverified claims do not.

    The shared Start coordinates game specialists and checked public learning. Private blueprints use your existing encrypted owner vault.

    Unique assets remain in checked cloud storage. Compact procedural recipes do not eliminate storage or establish game-production quality. Public AI uses released evidence only.

    Vehicle release targets

    2,000 vehicles for launch; 1,000 per monthly update up to 100,000. These are production targets, not finished or approved vehicles.

    Vehicle authoring programme

    Original car, aircraft and spacecraft profiles connect to your Etheraio Design, Image and 3D stages.

    Profiles are authoring targets. Finished, drivable vehicles require geometry, rigging, gameplay and production checks.

    Automatic game production

    Choose the existing Unreal game once; Start mining runs supported asset integration automatically.

    Generated assets enter the game staging queue. Saved jobs resume after restart. Full-game release requires passing gameplay, performance, rights and online-service checks.

    Automatic training in this mine

    One Etheraio AI coordinator connects specialist tools and checked evidence. Model weights stay architecture-specific.

    Hardware training checks wait for Start.

    Original paired data is checked automatically.

      Start automatically runs ready medicine, Qwen, TRELLIS and supported specialist adapters. Device tests cover installed ROCm, CUDA, Intel XPU, Metal or DirectML runtimes; CPU fallback requires your CPU permission. New candidates stay separate from your working models. Missing tools or data appear in the list above. New specialist adapters and generated training pairs remain experimental and do not automatically create funded cash entitlements until a separate network-verification contract supports them. Synthetic speech and the original small motion rig do not establish human voice quality or native character-model compatibility.

      3D training scope and limits

      The optional generator setup trains separate adapters through all three pretrained stages (3.88 billion base parameters). The base weights stay frozen. Original vehicle controls use the zero-image branch; image-guided training still needs licensed DINOv3 access. Your RX 9070 XT completed the three-stage checks using AMD ROCm. Other devices need compatible verified tools and enough admitted memory. CPU-only devices keep the tested component-training fallback. Numerical training checks do not approve game art: a generated mesh failed topology checks and remains unaccepted. Smaller component and material-head training are retained when the larger setup is unavailable. No automatic download or model replacement.

      Offline media specialists await their compatible models and runtime setup.

      Specialist coverage and next integrations

      One Etheraio interface coordinates separate specialist models. Each keeps its licence, model files and evaluation. Training candidates must pass their own checks before replacing working models.

        Characters, movement and faces

        Kimodo is selected for body animation, ARDY for responsive movement, and Audio2Face-3D for faces and lip sync. The optional 2.71 GB character package connects checked face inference and unconditional body movement to the shared Start. It has executed on CPU and this RX 9070 XT. Text-directed motion still needs the separately licensed Llama encoder; weight training still needs licensed paired data and benchmarks. Generated motion is prototype evidence until rigging, contact and game checks pass.

        Animation checks and hardware

        Each specialist keeps its own licence, candidate weights and evaluation behind Etheraio AI. Body motion uses a checked CPU or compatible PyTorch GPU path, including the tested AMD ROCm adapter. Face geometry uses a separate compatible ONNX path; this does not run NVIDIA’s CUDA-only facial SDK through ROCm. Skeleton, contact, retargeting and actual game checks must pass before production acceptance. Model selection is a starting point, not a measured claim that it is the best. Voice, face and training-data rights remain required.

        My model adaptation workspaces

          A prepared workspace records the original source, exact base revision and training requirements. It does not mean that its weights have been changed. The existing Qwen3-0.6B public trainer executes with Start when compatible tools and checked data are ready. Offline media adapters run only when their pinned models and compatible runtimes are configured. Other weight trainers still need their own licensed datasets and tested training tools. No automatic large downloads, paid APIs or private-data sharing.

          Etheraio coordinates separate specialist models; incompatible model weights are not merged. Recomputed numerical evidence improves later model selection within the same reference family and model digest, after eight distinct examples. Recheck / reselect to apply updated choices. Medicine uses scientific engines. A language-model answer cannot approve a medicine or certify a game asset. Existing full-block reward checks remain in force.

          INSTALLED MODELS · LOCAL ONLY

          Local AI · part of your mine

          Checking connected training…

          Detecting suitable work segments…

          Check your installed Ollama models.

          No models checked.

          Leaves room for Windows and other apps. Your manual memory budgets remain saved while automatic allocation is on. GPU-only work still uses system RAM and a host thread to stage models and results; it does not enable CPU model layers. Choose Maximum to also allow CPU-only science and preparation work.

          Include local AI above, save controls, then press the shared Start button. CPU verified work continues while the GPU runs approved material jobs and batches of distinct science references. An 8,192-token context permits up to 16 references per run; the largest supported context permits 32. CPU-only batches stop at four. Every candidate is checked separately. The model proposes numeric candidates; separate checks recalculate them. Passing network references add verified work weight when that policy is enabled; local-only references do not automatically create funded cash entitlements. Reference success is not a discovery. The mine routes suitable installed, licence-reviewed Qwen3/3.5, coding, Qwen-derived DeepSeek and OLMo families. Hardware detection sizes reference segments and prompt micro-batches. Compatible model layers can use GPU plus system RAM or CPU alone; the complete model still needs enough admitted memory. The runtime checks actual GPU residency. Maximum and Optimal favour larger variants that fit. Short palette jobs use a smaller context to leave room for the model; science batches retain their selected context. The chosen model stays pinned for its work sector during this app session as memory changes. Recheck / reselect lets you choose automatically again. A smaller device can run the same compatible model with CPU layer offload and smaller active segments when enough system RAM is available. Otherwise that stage waits for capacity or needs a smaller model. Ollama controls allocation; this budget is not a hard VRAM cap. No shell commands, model downloads or remote models. Generated text needs review; private medical simulations use scientific engines. Local AI jobs do not automatically create funded cash entitlements.

          AI output speed
          Unavailable
          Model / backend
          Unavailable
          Model VRAM
          Unavailable
          Local jobs · completed / failed
          0 / 0

          No AI reference candidates evaluated.

          Last local AI candidate and checks
          No job completed.

          ORIGINAL GAME ASSETS · LOCAL GPU WORK

          Game assets · part of your mine

          The first six concept designs are Auvryxis Aster S, Veyra R, Meridian GT and Orison, then Norzavique Summit and Forge. Orison and Forge are electric concepts; the other four use combustion engines. Their original emblems and factory colours are included. Sixteen original car families include sports coupes, supercars, hypercars, grand tourers, saloons, hatchbacks, estates, luxury saloons, crossovers, SUVs, off-road vehicles, pickups, vans, city cars, track cars and rally cars. Custom seeds vary the bounded proportions and available AI paint palettes. Start with an original car, then a georeferenced Earth coastline tile, alternating through a bounded 648-tile global grid. Generate PBR maps, UVs, normal/AO bakes and Blender previews; supported Unreal checks run afterwards. Earth uses coarse public-domain Natural Earth coastlines on a WGS84 ellipsoid. Detailed elevation, streets, buildings, rigged driving and photorealistic acceptance remain later work. Include GPU game assets above and choose Maximum for high-quality 4K maps on devices with enough VRAM. Start mining also starts this worker; you do not need to stop mining.

          Original car families use road-car proportions with distinct coupe, saloon, hatchback, estate, SUV, pickup and van silhouettes, fitted lights and glazing, panel seams and detailed wheels. Seeds vary each design and paint palette. These are static exterior studies; interior, driving and final art acceptance remain required.

          Auto measures permitted CPU/GPU render modes after Start; calibration earns no contribution weight.

          Jobs are off. Check installed Blender.

          No asset job completed.

          Blender 5.x is required. Supported asset jobs automatically run UV/geometry checks, height-driven vertex material binding, visible render comparison and a Windows content cook when licensed Unreal Engine 5.8 is installed. Packaged runtime, photorealistic art and licence rights remain separate gates. Local review is evidence, not independent production acceptance. AMD HIP has been tested on this PC; NVIDIA OptiX and Intel oneAPI require compatible installed drivers and have not been tested here. AI palettes retain model and output provenance. Original cars and coarse Earth coastline tiles are prototypes, not a completed full-Earth game; a proposed new material needs a separate validated simulator before scientific acceptance. Texture generation and PNG compression run in parallel within your selected CPU thread budget; baking and rendering use GPU. Checked cloud archives release only matching managed temporary sources; engine backup logs stay on this PC. Battery and application pause rules apply. The shared Stop button cancels every worker. Cinematic jobs use native 8K source maps, native 8K normal/AO bakes, an 8K UHD preview and up to 2,048 adaptive render samples. Hardware usage varies with scene size. Auto measures allowed CPU, GPU and combined modes on a bounded test scene and caches the fastest completed result; actual asset timings can differ. Both CPU and GPU permissions are required for combined mode. Tiled rendering retains the chosen source detail; CPU-only rendering is available when allowed and sufficiently provisioned. 8K source resolution is not proof of photorealistic production quality. Local output does not automatically create funded cash entitlements until independent game validation and network verification exist.

          Medicine & private researchYour private vault, medicine ideas and simulation tools.

          OWNER ACCESS · EXISTING KEYS

          Your private chemistry vault

          Sign in inside this app with your original Etheraio owner account. Browser and app sign-ins are separate. Your existing Google vault is on www.etheraio.com; contribution billing is on etheraio.com.

          1. Restore older access and confirm the app’s recovery dialog.
          2. Copy your current key, then open your chemistry records.
          3. Paste into Key to unlock saved chemistry records and choose Unlock record.

          Use the existing key on this PC. Recovery does not create a new key or replace your encrypted records.

          Copy places only your key on the clipboard for 60 seconds. Keep it private.

          Recovery works with earlier keys still protected on this Windows profile. Missing keys without a protected copy cannot be recovered. Medicine workers can stay off while you restore access.

          OWNER RESEARCH · PRIVATE INPUTS

          Screen my medicine idea

          Queue one molecular structure for the eleven installed numerical models, alerts and conformation checks under the shared Start. Your research condition is a hypothesis; it does not establish that the molecule treats it.

          Owner sign-in and existing protected medicine setup required. Inputs stay protected on this Windows profile; reports upload encrypted to your owner vault. Queuing does not start mining.

          No patient details, doses or synthesis instructions. These checks predict selected properties and hazards. Disease efficacy, many organ-specific endpoints and clinical approval remain unavailable. Private screening does not automatically create funded cash entitlements.

          MEDICINE · PART OF YOUR MINE

          Medicine checks in your mine

          Start mining runs the enabled public reference checks with the rest of your work. These checks use known public molecules, structure and conformation calculations and eleven installed numerical models. They do not need an owner login or vault key. Native tools must be installed; public reference reports are saved in your checked account cloud outputs.

          These model references currently do not automatically create funded cash entitlements because their network reward acceptance is not configured. The mine automatically records endpoint coverage, catalogue flags, conformation evidence and model-domain assessments. Missing, inconclusive and outside-domain predictions remain labelled; no prediction approves a medicine. Existing eligible network work keeps accumulating equal reward-block weight.

          Shared medicine tool locations

          Use installed Python with RDKit, Java 17 or newer and the audited VEGA 1.2.6 JAR. These paths contain no vault password.

          Optional owner-only unpublished chemistry

          Private molecule screening is an additional mine worker for the pinned owner account. Unpublished structures never go to public miners.

          Prepare the protected owner setup once if you want private candidate checks.

          A new key is generated on this PC and protected by Windows. Restore older vault access lets your current key open records with recoverable earlier keys. Original encrypted reports and key copies stay preserved. Start/Stop controls both workers.

          Open the vault access controls at the top ↑

          Private candidate screening is off

          Loading saved medicine progress…

          My chemistry simulation reports ↗ · My cloud outputs ↗

          Owner-only reports stay in your encrypted lab. The expanded 11-model suite screens the same 36 bounded inputs once. Earlier four-model records stay preserved; the extra checks have their own completion count. Finished suite batches do not repeat; unavailable scientific checks remain unknown. Clinical approval and qualified reward acceptance are not configured for these private simulations.

          MEDICINE FIRST · REPRODUCIBLE METHODS

          Scientific simulations in the shared mine.

          Start runs compatible public RDKit conformer checks and the approved OpenMM protein-control campaign when activated. A different account and Node must reproduce each result. A single sensitivity follow-up is queued after agreement; completed controls do not loop. These method checks do not automatically create funded cash entitlements and do not assess medicine efficacy.

          Automatic fixed-profile verification waits for activation.

          Fresh consented component checks, including failures, join continuous local learning under Start. Duplicate evidence is ignored. Compatible text specialists can train from these scoped facts; this does not retrain every 3D/voice model or establish better visual quality. Candidates stay separate from working weights until their own evaluation passes.

          Scientific engines are being checked.

          Offline editions use bundled RDKit/OpenMM files with the checked portable Python. The small edition may download up to 37.5 MB. Existing tools, thermal protection and resource limits stay preserved.

          ONE FARM · OPTIONAL SPARE CAPACITY

          AI first. Crypto when capacity is free.

          Use the same Start and Stop controls. Checked CPU and GPU lanes can run alongside AI work on measured spare capacity. AI work takes priority. Crypto uses measured capacity below the CPU or GPU limit you selected. Fresh readings and spare RAM/VRAM are required; crypto pauses as load reaches the target, and new GPU stages reclaim their memory first. This feedback target can briefly overshoot and is not an exact hardware cap. AI, game and science work always take priority. Leave the app running for continuous work, or choose lower CPU/GPU utilisation target. Pause, battery and application rules still apply; there is no fixed utilisation guarantee.

          Included estimates must use measured power, external fees, withdrawal costs and a device-matched worker. No guaranteed profit or payout. Sustainable-operation guardrails: AI/game/science work has priority, stale or missing quotes stop automatic coin selection, and actual block receipts alone back native payouts. Loss-mode mining can still cost the consenting user money; Etheraio’s overall financial sustainability cannot be guaranteed without ongoing settled income covering operating and reward costs.

          MEASURED AI RATE · COST, NOT A REWARD

          AI speed & profitability calculator

          Local model rate: No local model rate measured · custom rate: — ACU/s.

          One Etheraio Compute Unit (ACU) equals 100 output tokens from the current model. It is a speed label only, not a cross-model benchmark, reward weight or cash value.

          Estimated AI electricity/day
          Unavailable
          Scenario net/day
          Unavailable

          No income or profitability is assumed. Points and unspent AI credits are not settled cash.

          Crypto mining only

          Use only the checked cryptocurrency workers within your saved limits. AI, research, game work and contribution rewards are skipped in this mode. Monero pays in XMR. Pearl pays in PRL when a checked route becomes available.

          Crypto is stopped.

          Monero · CPU

          Pearl · GPU setup pending

          Spare-capacity auto: AI work takes priority. Checked Monero and Pearl routes follow your combined CPU/GPU targets.

          Confirmed earnings appear in the coin mined.

          Pool and device setup

          AI work has priority. Automatic comparisons use only checked, wallet-ready routes with fresh device, fee and electricity estimates. Other coins need their own receiving addresses and runtime checks. Native mode can cost more in electricity than it earns.

          Public P2Pool group mining pays directly in XMR with no Etheraio fee. Your address is public on P2Pool. Solo mining awaits a checked worker and payout route.

          The independent P2Pool CPU implementation can use the checked pool program without XMRig. Selection requires your confirmation and a synced node. It does not clear XMRig’s hold or enable a GPU miner. Group crypto settlement remains separate.

          Monero: sync status appears here.

          Pearl prepares its checked GPU worker and direct PRL payout route under Start.

          Public P2Pool uses temporary worker files. If a scratch check reaches 1 GB, Etheraio stops the worker, clears its scratch folder and starts a new session when it can.

          Sync nodes only validates public chains using permitted CPU work; it starts no AI jobs, miners or payouts. Stop and app close stop owned nodes. The node must retain chain data during sync. Choose a bounded pruned cache, or clear it after Stop and resync next time.

          After signing in, save your receiving address and pool. Start mining uses the spare CPU allowed by your settings once Monero has synced.

            CPU · Monero (XMR) receiving wallet

            Enter your public primary Monero address here. No website wallet setup or existing crypto balance is required. The address is saved for your signed-in account on this PC; never enter recovery words or private keys.

            You can type your address now. Stop the farm before saving. Saving sends no funds and does not start crypto.

            Crypto payouts and cash contribution income are separate. The same coin mined is paid to this receiving wallet only after real pool blocks are found and mature. Runtime checks, a synced node and separate crypto permission are still required.

            MEASURED THROUGHPUT

            AI farm & crypto rates

            AI output · this device

            No measurement

            Generated language-model tokens over the last 60 seconds, including idle time.

            CPU · Monero

            Stopped

            Only measured RandomX hashes are shown.

            GPU · selected coin

            Stopped

            The selected coin needs a fresh reading from its own worker.

            Shared AI network

            Unavailable

            Opted-in device reports; different models count tokens differently.

            Images, models and scientific results keep their own job counts. Tokens and hashes are performance measurements, not accepted weight, Points or confirmed crypto earnings.

            Use the receiving address already saved on this PC, or import your saved website address when available. You will confirm it before use. No funds are sent; checked tools and a synced node are still required.

            Wallet and pool setup
            1. Create receiving wallets for the selected coins.
            2. Connect checked CPU and GPU pool routes.
            3. Prepare compatible, verified mining tools.
            4. Measure net earnings after electricity and all fees.
            5. Connect and test confirmed native-coin payouts.

            The intended split is 99.9% of confirmed pool receipts to contributors and 0.1% to Etheraio, after external costs. Users pay their withdrawal fees. Payouts keep the exact mined coin and network; there is no exchange, swap or conversion into Points.

            Keep wallet creation and recovery words private. Enter no seed phrase or private key in this app. A user-operated pool node, checked miner and exact 0.1% fee settlement still need preparation; this panel does not create them. The optional Etheraio P2Pool variant splits native block rewards between miners and the owner. It uses a separate peer network; its initial capacity and payout frequency are not established. A farming client alone is not a synchronised blockchain node.

            A completed AI calculation is not a cryptocurrency proof. Public hashes can seed procedural visual variations, but the hash alone earns no game weight or extra reward. Crypto stays off in the Microsoft Store distribution.

            GPU COIN PREPARATION · NATIVE PRL

            Pearl · public receiving address

            Save your controls, then use the existing Start mining button with optional crypto consent. Missing or blocked workers stay waiting while ready AI and Monero work can continue.

            Pearl uses the pinned BzMiner worker and a checked Kryptex TLS route. Start prepares its official download and a normal Windows Security scan, then monitors actual shares on the selected device. AMD, NVIDIA and Intel support depends on the specific GPU and driver.

            Only public receiving addresses belong here. Keep recovery words and wallet passwords private. This setup starts no mining or transfers.

            Direct Pearl route: 3% Kryptex pool fee plus 2% miner developer time; minimum automatic payout 1 PRL. The pool pays your saved PRL address, with no conversion and no Etheraio fee. Etheraio/group fee settlement is a separate future route.

            Optional and off by default. No receiving address, keys, prompt, files or payout amount is uploaded by this choice. Save controls to apply it. The website needs its compatible tracker update; turning sharing off clears its next heartbeat report.

            Monero CPU sharing waits five seconds after fresh measured spare-capacity permission. Otherwise the existing thirty-second idle grace applies. Fresh readings, your limits and AI preemption remain required.

            NATIVE COINS · SEPARATE FROM CASH CONTRIBUTIONS

            Your crypto earnings

            Sign in to read your own native payout records.

            Monero · Cake Wallet

            Recorded mature payouts
            Not verified
            Recorded payouts still maturing
            Not verified

            No earnings are inferred from hashrate.

            Pearl · PRL

            Worker-reported hashrate
            Unavailable
            Recorded mature payouts
            Not verified
            Recorded payouts still maturing
            Not verified

            Pearl hashrate appears after the selected checked worker starts.

            A verified PRL payout feed is not connected. Hashing and a saved address are not earnings.

            This tracker does not control your wallet, request keys, start mining or send withdrawals. Native payouts depend on real accepted blocks. Your wallet is the authority for its balance and spending.

            Node & storageNode registration, optional hosting and saved outputs.

            YOUR PC ON THE NETWORK

            Make your PC a node

            NOT SET UP

            Join Etheraio using this computer. Your Node is linked to your account, with its identity encrypted on this PC.

            Sign in to set up your PC Node. Registration leaves mining off.

            Registering a Node does not start computing. Choose your CPU/GPU utilisation target and press Start mining when ready. Start can also register this PC automatically.

            OPTIONAL WEBSITE FILE HOSTING

            Make this PC a node and server

            SERVER OFF

            Help keep approved Etheraio website files available. One click registers your PC Node and starts a file server through Cloudflare. Keep this app open while hosting.

            File server is off.

            50 GB is a limit, not an immediate allocation. The cache grows only as approved files are requested, leaving at least 2 GB free.

            This release hosts verified public images through Cloudflare. Accounts, private research, payments and application code remain on the protected host. No inbound ports or router changes are needed. File hosting earns no mining Points.

            TEMPORARY DEVICE STORAGE

            Cloud-first mine outputs

            New connected-mine reports upload privately before temporary cleanup.

            Assigned assets and AI reports are uploaded, downloaded and hash-checked before their managed temporary copies are removed. Failed uploads retain their copy and pause new generation when backed up. Working models, small receipts, installed tools and existing files stay on this PC. Research privacy and game acceptance still apply.

            File assistanceChoose files and review proposed edits before applying.

            LOCAL FILES · YOUR EXPLICIT CHOICE

            Create with Etheraio · local file assistance

            Connect one project folder for bounded local search, or choose up to three specific 4 KB files. Project prompts read at most six relevant text/code files (16 KB each, 24 KB total). Only those selected files are sent to your installed local model. No shell commands, web uploads or shared training. Review the proposed replacement before applying; a protected original backup stays on this PC.

            No files selected.

            HTML proposals include a static visual preview; scripts, links and external assets are disabled. Other files show a code preview. Game projects and scripts are not executed or tested here.

            Activity & system detailsAccepted work, application rules and device information.
            AUTOMATICALLY MAKE SPACE
            When another app is in use
            Choose an application and the pace to use while its window is active. Return to your chosen pace when you switch away.
            Choose a game executable to pause farm work while its window is active. Your selected farm pace resumes after you switch away. Closing Etheraio stops its workers; these controls do not change Windows power or graphics settings.
            Application names stay on this PC. These rules never start mining on their own.
            DIRECT ACCESS TO YOUR HARDWARE
            Your system
            Processor
            Detecting…
            Memory
            Detecting…
            Graphics
            Detecting…
            Processing engine
            Native CPU / DirectX 12
            To view GPU activity in Windows Task Manager, change a GPU graph to Compute; the 3D graph may stay idle. Live activity counters appear above; temperature monitoring is unavailable. No overclocking, driver installation or Windows power changes.
            GAME DEVELOPMENT TARGET
            25 October 2030 · four-year target
            No completion forecast yet. A full production workload and measured accepted-game throughput are required.
            The four-year date is a planning baseline for one PC running continuously. Once its sustained production-accepted game rate is measured, the network capacity estimate moves earlier or later with accepted game throughput. No accepted throughput pauses the capacity clock. Quality, content and platform checks still determine release readiness.
            ACCEPTED WORK RATE
            0 verified jobs / minute
            No job processed. Everyday presets leave the GPU off.
            Verified result reports appear automatically after a block. Reused layouts still require art, physics and performance review.
            My result reports ↗
            Job timing appears after verified work.
            GPU only always uses compatible selected GPUs. All compatible GPUs splits candidates or samples across devices. Optimal compares permitted GPU and CPU paths and selects 80% CPU / GPU utilisation target when clicked. This is a balanced starting point with headroom, not a measured earnings or hardware-life optimum. Existing saved controls and selected quality stay unchanged until you choose a preset. Maximum requests up to 100% effort. Compatible language-model batches grow from measured speed and available VRAM; memory-heavy GPU stages stay serialized while independent CPU work can overlap. Actual utilisation varies by task and runtime; no setting can force a constant 100%.
            Sign in securely
            Use Google, supported ChatGPT sign-in or email in your browser, then confirm the matching app code. Your password stays in the sign-in page.
            Network AI processing
            Offer this PC as a reviewed network AI node. Requests use the same pinned 27B model and your saved mining CPU/GPU and memory settings. A compatible installed model, enough memory and operator review are required. Customer text is excluded from shared training. Processing nodes can read assigned text; other nodes cannot read it. Private files are never attached.
            Shared processing is off.
            Connection does not start work. Once reviewed, Start mining serves available requests through the shared scheduler. Stop cancels processing and releases its request. A node's reported token counts do not automatically earn verified contribution weight.
            PAID LOCAL SOFTWARE · YOUR ACCOUNT
            Your licence & automatic updates
            Restore a verified purchase using this app's sign-in. Licence files are signed by Etheraio and saved using the operating system's protected key store. Existing free experimental functionality is preserved.
            Licence delivery is loading.
            Licence
            No active paid licence
            Local use
            Restore a verified purchase
            Eligible updates through
            Not activated
            Next release-window check
            Next available window
            Two London calendar release windows each month: on/after the 1st and 15th while this app is open. Only passing, signed and licence-covered packages are offered. Missing candidates stay waiting. Permanent licences retain eligible software use after three years of updates; monthly access ends with the paid period. Hosted/grid compute is separate. Windows protection stays enabled.
            No checked update selected.
            

            DigiDash GridShare public hub documents over existing local links.
            SELECTED NETWORK SHARING
            Your hub. Your bandwidth.
            DigiDash Grid links explicitly paired local hubs. DigiDash Broadband Share is the separate planned service for sharing an authorised internet connection. Compatible radio hardware and real path tests determine reach and speed.
            Grid is off.
            No public hub is running.
            Stop mining before changing roles. Pair both devices using their public codes. HTTPS checks the chosen certificate and signed peer identity. Existing local connectivity is required; this does not change Windows networking or extend a radio’s physical range.
            Public payload copies use bounded RAM and expire after five seconds; a small protected identity and configuration persist for reconnecting. AI work has priority. CPU scheduling is approximate; the hub pauses when current headroom is unavailable. Accounts, passwords, reset links, payments and private research stay with their protected services.
            DigiDash Network Units (DNU) measure independently accepted network delivery. Local counters, pairing and tests create no cash earnings. Paid broadband and internet forwarding remain unavailable until their separate radio, gateway, delivery and provider checks pass.