Show HN: DeepClause – A Neurosymbolic AI System Built on WASM and Prolog (github.com)

DeepClause is a new neurosymbolic "missing logic" agent that marries Prolog-style symbolic reasoning with LLMs via a Prolog-based DSL called DML (DeepClause Meta Language). Announced as a pre-release desktop app (Electron) with WASM-built SWI‑Prolog and an embedded v86 Linux VM, DeepClause encodes agent behavior as executable, typed logic programs rather than fragile prompt engineering. Key primitives (e.g., @-predicates for structured prediction, system/ observation/ chat/ generate) let Prolog control flow, backtracking, and verification while delegating semantic tasks to configurable LLMs; the runtime verifies typed I/O, tracks full execution state with a custom meta-interpreter, and enforces strict sandboxing so community-shared DML skills are more auditable and safer than raw prompts. For AI/ML practitioners this matters because it tackles perennial LLM weaknesses—non‑determinism, poor multi‑step logical reasoning, and uninspectable decision-making—by making reproducibility, debugging, and symbolic integration first-class. Architecturally notable features include hybrid memory (separating conversational memory from execution state), native support for backtracking and constraint solvers (CLP(FD)/CLP(R)), and a compile‑then‑execute workflow that can auto-generate or run human-written DML. Early binaries are available but the core Prolog engine is currently distributed as a binary only; the project is experimental with performance and stability caveats (WASM speed, bugs), yet it offers a practical pathway toward accountable, verifiable agent systems.