langgraph-jev: developer wraps TypeSafe's Jev API into a typed LangGraph decision node with confidence-scored answers

Releases·Tool Release

jevcore: TypeSafe Jev typed-judgment library ships for DeepSeek Harness, MCP, and plain Node

PerryLink published jevcore on GitHub, a TypeSafe Jev integration exposing calibrated yes/no, choice, and score judgments to agents instead of prose. It ships as three packages — a standalone core, a DeepSeek Harness plugin, and an MCP server — and is offline by default with disclosed egress. It matters to practitioners building agent harnesses that need a narrow, typed decision surface rather than free-text model output.

PerryLink published jevcore on GitHub, a TypeSafe Jev integration exposing calibrated yes/no, choice, and score judgments to agents instead of prose. It ships as three packages — a standalone core, a DeepSeek Harness plugin, and an MCP server — and is offline by default with disclosed egress. It matters to practitioners building agent harnesses that need a narrow, typed decision surface rather than free-text model output.

Releases·Tool Release

LLM2Jev turns local LLMs into Jev-style structured decision engines with prefill-only binary inference

A new open-source project, LLM2Jev, adapts local Hugging Face-compatible causal language models into Jev-style structured decision engines, emitting typed Choice, Score, and Noul answers with probabilities. It uses prefill-only binary inference — scoring candidates with yes/no logit judgments without generating answer tokens — and leverages SGLang's Radix Cache for shared-prefix efficiency. The project is independent and unaffiliated with Jev or TypeSafe.

A new open-source project, LLM2Jev, adapts local Hugging Face-compatible causal language models into Jev-style structured decision engines, emitting typed Choice, Score, and Noul answers with probabilities. It uses prefill-only binary inference — scoring candidates with yes/no logit judgments without generating answer tokens — and leverages SGLang's Radix Cache for shared-prefix efficiency. The project is independent and unaffiliated with Jev or TypeSafe.

Traser is a debugging tool for multi-step LLM workflows built on LangChain and LangGraph. It helps engineers investigate executions that technically succeed but produce wrong answers, by letting them compare a suspicious execution against a known-good one. It targets engineers building multi-step AI systems who struggle to find which part of a trace actually matters.

notjev is a Node library that makes any OpenAI-compatible model produce fast multiple-choice decisions. It requests one token with logprobs enabled, renormalizes the letter probabilities, and returns a verdict, probability, margin, and abstention threshold. This gives Jev-style Choice/Score/Noul outputs to developers running local models (vLLM, llama.cpp, Ollama) without runtime dependencies.

Products·concept

Framework-agnostic intent & tool router for agent loops using TypeSafe's Jev decision model

A lightweight router that uses TypeSafe's Jev model to decide tool calls and agent paths quickly. Jev returns typed Choice and Score primitives instead of generated text, cutting latency and cost versus generative LLM routing. Aimed at developers building low-latency agent architectures.

A lightweight router that uses TypeSafe's Jev model to decide tool calls and agent paths quickly. Jev returns typed Choice and Score primitives instead of generated text, cutting latency and cost versus generative LLM routing. Aimed at developers building low-latency agent architectures.