Decidealot
Serves local decision models over TypeSafe-compatible HTTP and MCP, converting raw state into structured choices, scores, and yes-no calls.
Decidealot empowers users to deploy and run Laya and Von local decision models directly on their own hardware, eliminating cloud dependencies and associated costs. It processes messy input states into clearly typed outputs like choices, scores, and probabilistic yes/no answers. Designed for efficiency, Decidealot downloads and verifies model bundles at startup, loading specific models on demand and intelligently unloading them after an idle period to conserve resources. It offers a standardized API and MCP Streamable HTTP interface for seamless integration into various applications and agent workflows.
Características Principales
01Local model serving for Laya and Von decision models
02TypeSafe-compatible HTTP API and MCP Streamable HTTP
03Automatic model loading and unloading based on idle time
04Containerized deployment with Docker, including CUDA support for GPU acceleration
05Support for typed choices, scores, and yes-no (noul) answers with probabilities
061 GitHub stars
Casos de Uso
01Integrating local decision intelligence into agent workflows
02Automating policy, routing, triage, or approval decisions
03Evaluating different decision models (Laya vs. Von) for specific use cases
04Self-hosting AI inference to avoid cloud bills and ensure data privacy