typesafe-ai

What problem does it solve? Turning natural language understanding into reliable application logic usually means fragile LLM prompt-and-parse pipelines. This Skill helps you replace those with typed judgments and probabilities from TypeSafe's System One models (like Jev), so your code owns the workflow while the model supplies programmable common sense. ## Core Features & Use Cases - Typed Primitives: Design judgments using Choice (pick one option), Noul (probability a condition holds), and Score (degree along a dimension) primitives with structured state, instructions, and criteria. - Composition Patterns: Apply proven patterns like routing with argument filling, reranking, value extraction, citation verification, composite scoring, and speculative fan-out, guided by live docs and cookbooks. - Uncertainty Handling: Use probabilities and confidence thresholds evaluated on your data to route, escalate, or verify decisions. - Use Case: Build a support ticket router that classifies intent, extracts typed parameters, scores urgency, and escalates low-confidence cases to a human, all as composable typed judgments rather than free-text prompts. ## Quick Start Ask the assistant to design a TypeSafe integration that routes incoming user requests to typed handlers using the Choice and Noul primitives, reading the live docs at docs.typesafe.ai first.