Skip to content

Quickstart

Fabrizio Degni edited this page Nov 16, 2026 · 5 revisions

First customer Hive run

Open NodeLift Account with your private pilot invitation. Create the account, obtain actual operator approval and sign in with your chosen session duration. Set up account Authenticator before device management.

Select your personal cluster, install a compatible companion, create the scoped QR and scan it. Enter the private code shown by that device in the portal. Save operating limits, prepare the matching model and enable participation on the companion, then run a first supported task from the wizard. No Mac inference host or relay is required. See Customer onboarding for recovery, limits and validation status.

The operator Hive manages the separate owner pilot. Existing operator enrollments cannot be silently adopted into a customer workspace. Logical clusters do not establish shared model memory. The bounded pilot engine guide separates actual Linux evidence from pending phone qualification.

Development-only local harness

The following steps reproduce historical local controls. They are not the product onboarding path.

Start with the computer alone. On Apple silicon, the standalone host preview includes Python, dashboard assets and the native inference library. Extract it and open Start NodeLift.command. It checks the host prerequisites and opens a private local dashboard; a GGUF model is still needed for inference. It is ad-hoc signed, not notarized, and has been tested on one M3 Max.

Build from source

The source build requires Python 3.12+, Git, CMake 3.22+, a C++17 toolchain and OpenSSL. The build command fetches the locked llama.cpp revision and compiles it locally. Another OS or backend needs separate qualification.

git clone https://github.com/sev7enITA/NodeLift.git
cd NodeLift
python3 scripts/doctor.py
./scripts/build-native.sh
python3 scripts/fetch-control-models.py

The last command downloads about 1.61 GB of public control weights. Read the pinned model inventory and the models’ licenses first. It is optional if you already have suitable complete GGUF files.

  1. Find the computer’s private LAN IPv4 address on the network shared with the phone. An address such as 192.168.1.20 below is only an example; substitute your actual address.
  2. Start the host:
PYTHONPATH=host python3 -m nodelift.host --address 192.168.1.20
  1. Open the private dashboard link printed by the host. The plain http://127.0.0.1:9868 address loads the page but does not grant control access.
  2. In Models, import a complete GGUF or register an existing absolute file path. For the control pair use Qwen2.5 1.5B as target and 0.5B as draft.
  3. Select the target/draft in LLM engine, apply configuration, choose Computer only · baseline, enter a public test prompt and run 32–64 output tokens.
  4. Only after this succeeds, follow the Android or iOS guide.

Choose Language → English / 简体中文 / Italiano. The choice is saved locally. There is no required cloud account. OpenRouter is an optional, explicitly selected cloud backend with possible charges.

Expected result: actual generated text and a completed run with timings. Compilation, a green node indicator or RAM telemetry alone is not proof of accelerated inference.

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.

Clone this wiki locally