Laya

Laya is an Apache-2.0 Python package and encoder-model decision engine developed by Convai Innovations. It accepts text or JSON state and evaluates typed questions in a single forward pass, returning categorical choices with probabilities and confidence, ordinal scores, or calibrated Boolean probabilities rather than generated text. Its stated uses include routing, ticket triage, prompt-injection detection, moderation, phishing detection, and churn-risk classification. Laya includes English, multilingual, and typed-decisions checkpoints, plus a router that selects a checkpoint using explicit settings, detected script, and best-effort language detection. The checkpoints use reinforcement learning against strictly proper scoring rules (RLCD); the project also provides fine-tuning and temperature-calibration workflows. The multilingual checkpoint is intended for more than 100 languages, while the repository cautions that the base models can perform near chance on some zero-shot typed-decision tasks and that large choice sets are constrained by the option-token budget.

Laya-CoreML is an open-weight on-device decision-model runtime and port of Laya for Apple Silicon, using Core ML and Apple's Neural Engine. It answers typed choice, ordinal-score, and boolean questions with probabilities or scores rather than generating text or JSON; its Python API loads a model and returns structured answers. The repository includes Core ML checkpoints, tokenizer and configuration files, an offline Python package, a command-line Snake demo, and conversion and benchmarking tools. The Neural Engine bundle uses a short fixed-capacity input path, while longer or general-purpose requests can use CPU-and-GPU checkpoints. The project documents a separate Neural Engine graph using BC1L activations, 1×1 projections, and per-head attention, with CPU handling input and output boundaries. It supports macOS 15 or later on Apple Silicon and can run without PyTorch, Transformers, or MLX at inference time. The project is licensed under Apache-2.0 and describes itself as an independent port by Convai Innovations and contributors, not an official Convai Innovations or Apple release.

Ollaya is an open-source system for running decision models locally. It accepts a message, email, ticket, or other JSON state together with typed questions and returns calibrated probabilities for choices, scores, or binary decisions rather than generated text. The Ollaya daemon pulls models by name and serves them through a CLI and HTTP APIs, including TypeSafe-compatible `/v1/systemone`, `/v1/decisions`, and `/v1/models` endpoints. Its native API adds routing information and timings, while model definitions can embed reusable question sets. It supports ONNX Runtime on CPU and NVIDIA GPUs, and llama.cpp for GGUF models on CPU, CUDA, and Apple silicon. Ollaya also provides an MCP server for clients such as Claude Code, Claude Desktop, and Cursor. It publishes small ONNX graphs while obtaining the original model weights from their authors' Hugging Face repositories, pinned and verified by commit and SHA-256; it does not re-host those weights. The project is distributed under Apache-2.0, while individual models retain their own licenses, and it is available as a CLI, daemon, desktop application, and Docker image.

LocalJev is a TypeScript service that provides a local Jev-compatible decision API for applications using TypeSafe's typed decision interface. It exposes a POST /v1/systemone endpoint and translates typed Jev questions and application state into classification prompts for an OpenAI-compatible DiffusionGemma inference server. The service validates and retries malformed JSON, normalizes probability vectors, calculates Jev-compatible choices, expected scores, and entropy-based confidence, and returns the standard Jev response shape. Its probabilities are model-generated or self-reported rather than read directly from model logits, so the repository advises evaluating calibration for the intended workload. It runs with Bun and can use oMLX as the local inference server; the default LocalJev address is http://127.0.0.1:8080.

jeff is a self-hosted, MIT-licensed drop-in implementation of TypeSafe's jev System One API, powered by the GLiFormer encoder model. It accepts classification questions through a compatible SDK or the POST /v1/systemone endpoint and returns choice selections, ordered scores, and noul yes/no probabilities. The server supports CUDA, Apple MPS, CPU, and ONNX Runtime deployments, with batching, bearer-key authentication, rate limits, request limits, health and statistics endpoints, and deployment configurations for Modal. Its repository reports lower hosting costs than jev but weaker results on reasoning-heavy tasks.