MESHMESH
06 · ECOSYSTEM

JOIN THE MESH

NETWORK STATUS
OPEN BETA — Q2 2026
MESH is entering open beta. Early access is available to developers, organizations, and anyone who wants to put their own devices to work.
PARTICIPATION TRACKS
TRACK 01
DEVELOPER
Build AI-native applications and agents using the MESH API. Instant key delivery for verified organizations.
  • RESTful + webhook-driven API
  • SDK for TypeScript, Python (coming soon)
  • 16 workload types across image, video, 3D, audio
  • Runs on hardware you already own — no per-GPU cloud bill
  • Async job API with webhooks on completion
TRACK 02
DEVICE OWNER
Run AI on the machines you already own. Open a browser tab and the device joins your own group.
  • No installation required — browser-native
  • Supports NVIDIA, AMD, Apple Silicon, Intel Arc
  • Invite a second device with a link — VRAM pools
  • Models too big for one machine split across the group
  • Desktop app keeps a machine available in the background
ADD YOUR DEVICES →
TRACK 03
ORGANIZATION
Deploy MESH for your team. Custom SLAs, dedicated node pools, usage dashboards.
  • Org-admin RBAC with member management
  • Reserved capacity pools
  • Custom SLA tiers (<500ms priority routing)
  • Usage analytics and audit log
  • Enterprise billing and invoicing
TRACK 04
PROTOCOL CONTRIBUTOR
Help build the protocol, add new model types, or extend the SDK.
  • Full protocol documentation available
  • IWorkloadHandler interface — add any model
  • TypeScript + Python test suites
  • Contributor program (coming soon)
VIEW DOCS →
FAQ
Is my data private when using MESH?
Job inputs are encrypted in transit and only ever reach devices in your own group. Nothing is persistently stored on them — OPFS caches model weights only, not job payloads.
How is result quality guaranteed?
Results are validated by consensus — the same job is run on multiple devices and outputs are compared by hash. Jobs that fail consensus are automatically re-run.
What happens if a node goes offline mid-job?
The scheduler detects stale leases (Redis TTL expiry) within 15 seconds and re-assigns the work unit to another device in your group. Most users never notice.
Can I run MESH on a server, not a browser?
Yes — the node SDK supports onnxruntime-node for server-side execution. WebGPU backend is available in Node.js 22+ via Dawn. Full browser support remains the primary path.
Whose hardware runs my job?
Only devices in your own group — this browser, your desktop app, your phone, and anyone you invited with your link. If none of them are online, the job does not start.
What is the minimum hardware?
Any WebGPU-capable GPU with 4 GB+ VRAM running Chrome 113+ or Edge 113+. Lighter devices handle lighter workloads (BiRefNet, BK-SDM-Tiny). More VRAM in the group = more job types.