laya-php: self-hosted PHP SDK for typed text classification with Laya, a Jev AI alternative

Products·building

Weigh Swarm: research RAG pipeline with inspectable evidence and cited claims

A research RAG prototype that parses supplied PDFs or a research question, proposes claims, aligns quoted passages against parsed text, and produces a cited draft. LLMs handle planning, extraction, and synthesis, while a separate layer called Laya makes bounded judgments (relevance, methodology, contradictions, answer verification). It targets researchers who need to inspect the exact passages behind every claim rather than trust opaque model outputs.

A research RAG prototype that parses supplied PDFs or a research question, proposes claims, aligns quoted passages against parsed text, and produces a cited draft. LLMs handle planning, extraction, and synthesis, while a separate layer called Laya makes bounded judgments (relevance, methodology, contradictions, answer verification). It targets researchers who need to inspect the exact passages behind every claim rather than trust opaque model outputs.

Releases·Tool Release

stuntd: a local Jev-compatible server on Laya that learns from your own traffic (no API key needed)

A developer released stuntd, an Apache-2.0 local proxy for typed LLM decisions (choice, yes/no, score) that records traffic from a provider like the hosted Jev API, trains a small head on the frozen open-weights Laya encoder, and serves answers locally with fallback to the provider below a confidence threshold. It matters to practitioners because it cuts ~380 ms paid round trips down to ~22 ms local inference on closed-decision tasks, with self-promotion and demotion based on agreement rates.

A developer released stuntd, an Apache-2.0 local proxy for typed LLM decisions (choice, yes/no, score) that records traffic from a provider like the hosted Jev API, trains a small head on the frozen open-weights Laya encoder, and serves answers locally with fallback to the provider below a confidence threshold. It matters to practitioners because it cuts ~380 ms paid round trips down to ~22 ms local inference on closed-decision tasks, with self-promotion and demotion based on agreement rates.

Products·building

Jein - hosted API for the open Laya decision model, Jev-compatible endpoints

Jein is a hosted API service the developer built around the open Laya model after being unable to get a Jev account. Jev and Laya are decision models that return structured output with confidence levels per option, faster and smaller than LLMs. Jein exposes Jev-compatible endpoints so the Jev SDK works with it.

Jein is a hosted API service the developer built around the open Laya model after being unable to get a Jev account. Jev and Laya are decision models that return structured output with confidence levels per option, faster and smaller than LLMs. Jein exposes Jev-compatible endpoints so the Jev SDK works with it.

Releases·Tool Release

Laya: The Open-Source Jev Alternative That's 7x Faster and Runs Under 1GB of RAM

Laya: An Open-Source Jev Alternative Claiming 7x Faster Speed Under 1GB of RAM

Laya: An Open-Source Jev Alternative Claiming 7x Faster Speed Under 1GB of RAM