Run GGUF directly
Native llama.cpp inference with the best available platform backend. Models load lazily, requests serialize safely, and memory stays bounded to the residency cap.
sysone is a loopback System One gateway. It routes
noul, choice, and score calls
across hosted Jev, builtin GGUF models, and backends you already run.
$ sysone pull
verified qwen3-0.6b · 365 MiB
$ sysone up
listening at http://127.0.0.1:13900
Native llama.cpp inference with the best available platform backend. Models load lazily, requests serialize safely, and memory stays bounded to the residency cap.
Prefer hosted, prefer local, restrict either side, or let
auto use Jev first and the smallest available local
model as fallback.
Loopback by default. Credentials stay in the environment. Request state and answers never enter config, manifests, or daemon logs.
Pin an exact model when reproducibility matters, or let sysone order candidates by policy, availability, parameter count, and operator cost rank.
Builtin GGUF mode scores the model's complete first-token vocabulary
over constrained answer labels. Every answer reports
coverage and confidence. A general GGUF is
not presented as a calibrated Jev model; use trained System One
weights when calibration is part of your contract.
sysone pull
sysone up
sysone status --json
View source on GitHub