Annotations, AOP interceptor, autoboxer, function registry, model-router interface, context, standard library, semantic tool selection, and two-phase orchestration. The deterministic backbone of PrOOPt; runtime modules plug in concrete model engines.

LLM-powered persona state management for PrOOPt. Annotate a POJO with @Persona and @Trait; the LLM becomes the state transition function — resolving, evolving, and querying personas without a hand-written rule table.

Compile-time annotation processor for @PromptFunction: keyword-heuristic type-compatibility checks, single-brace {param} placeholder binding validation, generic-type capture before erasure, hard errors for unsupported generic shapes, ...

Worked, runnable PrOOPt examples in one module: a Legal Analyzer (mixed deterministic and bounded/elevated AI), an n×n Linear System solver, and a BODMAS/PEMDAS DAG solver. Each runs against a deterministic mock router — no API key ...

The prooopt:validate goal — semantic build-time validation of @PromptFunction methods. Scans compiled classes, diffs a SHA-256 validation cache, batches uncached methods to a semantic classifier, fails the build on INVALID (and ...

Cloud model adapters for the CLOUD_FAST and CLOUD_ADVANCED tiers: Anthropic and OpenAI, over the JDK HTTP client. Add this module to grant functions elevated AI authority.

Embedded local LLM inference via ONNX Runtime — zero external processes, inference runs entirely inside the JVM on CPU, no data leaves the process. Uses platform threads only, so it runs on Java 17+.