Skip to content

Models and providers

Fabrizio Degni edited this page Nov 16, 2026 · 2 revisions

Models and providers

The Hive model catalog and peer distribution guide describes the separate cloud-governed public/private artifact registry, verified HTTPS peer transfer, private storage-peer credentials and current desktop/server qualification boundary. Distribution benchmarks measure file preparation, network payload and scoped energy; they do not establish faster inference or pooled RAM.

The Models panel offers three local import paths:

Path How to use it Boundary
Existing GGUF Enter an absolute local path and choose Add path Registers without copying
Browser file import Select a complete GGUF and choose Import selected file Uses bounded upload chunks
Hugging Face Search a public repository, inspect files, review the model card/license, select a GGUF and download Pinned revision and SHA-256; restricted models require manual download

Import validates the header. Loading in the pinned runtime determines architecture support. Split GGUF sets, automatic safetensors conversion and model-supplied Python are not supported. Target/draft vocabulary IDs must have compatible meaning. The native verifier currently refuses recurrent/hybrid architectures that lack qualified state rollback.

Select NodeLift native · local cluster for NodeLift’s host/local/phone strategies.

For OpenRouter, expand External engines and cloud access, save a session key, refresh models, explicitly select OpenRouter and review the model’s rates before generation. Prompts leave the local network and charges may apply. There is no automatic cloud fallback; restart clears provider keys from host memory.

For Local API · oMLX / LM Studio, enter that server’s loopback URL ending in /v1, refresh its catalog and select a model. This route calls the external server for text generation. Its caching, loading and scheduling remain external; NodeLift does not add phone drafting to that route.

Ollama itself is not patched. Compatible clients can call NodeLift’s documented OpenAI-style text subset; API shape alone does not establish complete client compatibility.

API, SDK adapter and provider instructions · Repository reuse assessment

Clone this wiki locally