@infernetprotocol/liquid-agent
v0.1.94
Published
liquid-agent/v1 contract and runtime client for Infernet's temporal-policy family: CfC/LTC liquid neural networks (ncps) with a GRU baseline, trained and served through an isolated, version-locked Python subprocess.
Readme
@infernetprotocol/liquid-agent
The liquid-agent/v1 contract and the client for infernet-liquid, the isolated Python runtime behind Infernet's temporal-policy family (CfC and LTC liquid neural networks from ncps, with a GRU baseline).
import { buildBundle, verifyBundle, validate, LiquidRuntime } from "@infernetprotocol/liquid-agent";
const bundle = buildBundle(reviewedEpisodes, { consent, strategy: "by_session", seed: 0 });
verifyBundle(bundle); // schema + hashes + secret scan
const rt = new LiquidRuntime(); // needs `infernet model runtime setup --yes`
const res = await rt.request("train", { bundle_path, output_dir, family: "cfc" });Design, protocol, error codes and threat model: docs/liquid-agent.
Runtime dev loop (from runtime/): PYTHONPATH=. python -m unittest tests.test_runtime, python -m tests.make_fixtures, python -m infernet_liquid bench --out-dir /some/dir.
