Technology: Under the Hood

MESH routes your AI workloads to the lowest-latency, highest-throughput nodes automatically. The protocol is designed for adversarial environments: every result is consensus-validated before delivery.

Request path

How your job runs.

  1. Step 1Submit

    API call or SDK. The job enters the MESH queue.

  2. Step 2Match

    WorkMatcher pairs the job to the best available node in under 100 ms.

  3. Step 3Assign

    The node receives the assignment over a WebSocket push, with HTTP poll as fallback.

  4. Step 4Execute

    ONNX Runtime and WebGPU run inference inside the browser sandbox.

  5. Step 5Validate

    Result hashes are replicated across N nodes and checked against the consensus threshold.

  6. Step 6Deliver

    Verified output is returned, the job is marked complete and billing is settled.

Protocol Guarantees

  • Delivery SLA

    99.7%

  • Result validation

    Hash-based consensus with configurable replication.

  • Fault tolerance

    Automatic lease re-assignment when a node drops, in under 15 s.

  • Nonce proof

    Per-job nonces prevent replay attacks from adversarial nodes.

  • Epoch locking

    Lease epochs prevent stale result injection.

  • Node selection

    Latency, VRAM and utilisation scoring.

Compute & Capabilities Workloads

Model CatalogRun any published model on the devices you connect.Browse the catalog
Generative AI

Multi-modal synthesis: text, image, video and 3D. FLUX.2, LTX-2.3, Hunyuan3D.

Scientific computing

Monte Carlo simulations, molecular dynamics and mathematical modelling on WebGPU.

Data engineering

Vector processing, point-cloud transforms and large-scale feature extraction.

Media pipeline

Real-time 4K/8K upscaling, depth estimation, background removal, frame interpolation.

Execution Modes

Single
One node processes the entire job. Ideal for small models with the lowest coordination overhead.
Parallel (tiled)
The input is split into tiles, each assigned to a different node, and the results are merged. Used by upscalers on large inputs.
Sharded
The model is split across N nodes, each holding a subset of the weights, talking over WebRTC. Used by FLUX.2-klein, LTX-2, Wan 2.2, Hunyuan3D 2.1, TRELLIS.2-4B and ACE-Step.

Node Lifecycle

Your browser becomes a compute node. ONNX Runtime Web runs inference directly on your GPU via WebGPU: no driver install, no CLI, no server.

  1. Step 1Connect

    The node registers through the MESH SDK and declares its VRAM and compute capacity.

  2. Step 2Advertise

    It announces availability over WebSocket and enters the availability pool.

  3. Step 3Receive

    WorkMatcher pushes an assignment. The node holds a lease with a liveness TTL.

  4. Step 4Execute

    Inference runs in a Web Worker on ONNX Runtime. The GPU stays dedicated to the job.

  5. Step 5Submit

    The result hash and output are uploaded and the lease is released.

  6. Step 6Ready

    The node re-advertises. A heartbeat refreshes its liveness every 30 s.

System Requirements

  • Browser

    Chrome 113+ or Edge 113+ (WebGPU required).

  • GPU

    Any WebGPU-capable GPU: NVIDIA, AMD, Apple Silicon, Intel Arc.

  • Minimum VRAM

    4 GB (Tier 3) · 8 GB (Tier 2) · 16 GB+ (Tier 1).

  • Bandwidth

    10 Mbps+ recommended for peer-to-peer model transfer.

  • Storage

    2–20 GB per model set in the browser file system (OPFS).

  • Uptime

    No minimum. A device is usable whenever it is online.

Node Tiers

Tier 1 · 16 GB+ VRAM
Premium jobs · 2.2× rate
Tier 2 · 8–15 GB VRAM
Standard jobs · 1.4× rate
Tier 3 · 4–7 GB VRAM
Light jobs · base rate
Connect a device