llms.surf / README

README — llms.surf

local AI dispatch orchestrator · formerly Oracfit · public cut v3.5.0 · built in Saquarema, Brazil

// what it is

llms.surf is a local dispatcher that sits between your agent CLI and the models. It sends the 1% of tasks that need a brain to your expensive frontier model and the other 99% to cheap local ones. No result is accepted until a mechanical oracle — a real command, run on disk — proves it.

The rule in one line: nothing ships on a model's word — not even the judge's.

// quickstart — 2 minutes, no API key

The dispatcher boots local against the stub runner. Frontier keys are only added when a task earns them.

// the pipeline

• stage 1 — spec open: the oracle is frozen by sha256 at ring open. Goalposts don't move.
• stage 2 — dispatch: 1% frontier thinking / 99% cheap sweat, picked per task.
• stage 3 — oracle: gates read artifact content on disk. exit 0 is the only victory.
• stage 4 — incident: every failure becomes an incident; every incident becomes a mechanism. 106 in this cut.

// adapters — the lineup

Eight adapters plus a stub ship in this repository. If your CLI speaks text, it can speak llms.surf.

// this repo

Clone and run. License is proprietary (see LICENSE) — not claimed as open source. If you own the GPUs, the 99% is local and tokens on that path cost ~$0.