Skip to content
Fabrizio Degni edited this page Nov 16, 2026 · 21 revisions

NodeLift documentation

Shared computing resources and a longer useful life for existing devices.

NodeLift is building cloud-orchestrated infrastructure to democratise access to memory, compute and services. Created by Fabrizio Degni, it connects authorised resources so a device can access workloads beyond its own hardware capacity. AI is the first implementation track. The mission and product value explain the purpose.

Target architecture revision 2.0. The global cloud Hive manages identity, policy and regional discovery. Regional hubs manage workload queues, placement, leases and recovery. All computers, workstations, smartphones, tablets and eligible consoles join as peer workers. A Mac has the same worker role as other computers; cloud orchestration does not require a particular personal device.

Required model partitions extend one model across compatible worker memory domains. Independent requests, verified draft assistance and complementary services provide separate execution routes. Placement considers memory headroom, supported operators, network and queue costs, user policy and the appropriate recovery contract. The console reports resource coverage and useful contribution at node, hub and Hive scope.

The target design respects operating-system budgets, thermal conditions, local-time operating windows and data-residency rules. It does not physically add RAM to a device. Learn the roles and contracts in the architecture guide and interactive architecture explorer.

Hive execution authority

Hive owns the product request and execution lifecycle. The development Mac simulates hub components and test runtimes; it is not a required coordinator, main-model host or permanent verifier. The current bounded pilot dispatches complete-model LLM requests to ready workers. Customer work is scoped to the authenticated workspace and cluster; this does not pool model memories. The execution contract and migration gates define the required transition; local draft trials remain development evidence.

Your personal workspace

Start at NodeLift Account for the invitation-only customer pilot. Set up account Authenticator, select or create a personal cluster, install a compatible companion, scan its scoped QR and enter the private confirmation code from the device. Then choose operating limits and verify a supported first task. Follow Customer onboarding. The operator Hive and existing global enrollments remain separate.

Downloads and release availability

Desktop packages are identified by OS, architecture and execution backend. The Windows Node 0.7.1-preview.1 CPU previews are available for x86, x64 (Intel/AMD) and ARM64; physical Windows qualification remains pending. The separate private CPU/CUDA worker has Portable 0.6.2 and Update 0.6.1 variants. See Desktop packages for contents, requirements and checksums.

Published packages have their own version and platform requirements. The public companion release is 0.2.31 build 35; the Downloads section and mobile package guide describe the available artifacts, signing and installation requirements. The target architecture is not a claim that every depicted capability is included in that release. The interface language order is English, Simplified Chinese and Italian.

Purpose Guide
Understand the mission Mission and value
Explore the target system Architecture · Interactive diagrams
Understand cloud roles Hive and regional hubs
Understand resource measures Compute metrics and aggregation contracts
Download available packages Downloads · Mobile packages
Set up a published package Quickstart · Troubleshooting
Add a mobile worker Android · iPhone and iPad
Select model interfaces Models and providers
Build and contribute Development and workflows
Find artwork and attribution Gallery · About and credits · Acknowledgements

Release history · Project website

These pages are maintained in docs/wiki. Publishing the Wiki requires a separate Git push; editing these sources does not update the hosted Wiki.

Unified Hive resource overview

The Hive console brings actual hubs/groups, four explicitly planned regional zones, memory, profiled compute, historical useful work, energy coverage and devices into one Overview. Supported sensor collection is automatic with explicit missing coverage; four planned regions do not establish four operational sites. See the European edge AI requirements study for fourteen proposed evidence gates, inspected sources and remaining gaps.

Automatic supported sensor collection is available with measured and estimated coverage. Android 0.2.26/build 30 and Windows Node 0.7.1-preview.1 retain platform-specific physical qualification limits.

Direct Hive LLM pilot

Hive now owns bounded complete-model generation without a mandatory Mac coordinator. See the LLM engine, setup, physical Linux measurements and remaining mobile gates. Complete-model placement does not pool device memories.

Governed model distribution

Browse the public catalog or manage authorized workspace-private models from NodeLift Account → Models. The catalog guide covers immutable manifests, storage-peer credentials and verified HTTPS chunks. Actual file-transfer results distinguish WAN overhead, co-located sources, cache reuse and missing energy coverage. This is separate from model partitioning or shared inference RAM.

Clone this wiki locally