neurosym-ai 0.1.3

Neuro-symbolic guardrails for LLMs: rules + repair loops + (optional) SMT.

NeuroSym-AI

Neuro-symbolic guardrails for arbitrary information

Validate, sanitize, and enforce policies on text, JSON, and LLM outputs using symbolic rules with optional language-model-based repair loops.

---

Overview

NeuroSym is an information-first guardrail engine designed to enforce explicit, auditable constraints on unstructured and semi-structured data.

Unlike LLM-specific guardrail tools, NeuroSym operates independently of model providers and treats language models as optional adapters, not core dependencies.

It is suitable for:

• AI agents and tool pipelines
• Structured LLM extraction
• Compliance-sensitive systems
• Research in neuro-symbolic AI and AI safety

---

Key Capabilities

Input (Text / JSON / Tool Output) ↓ Deterministic Repairs (Offline) ↓ Symbolic Rule Evaluation ↓ Optional LLM Repair Loop ↓ Validated, Audited Output

Highlights

• Provider-agnostic (no model lock-in)
• Deterministic by default (no API keys required)
• Symbolic core (rules, schemas, constraints)
• Optional neuro-symbolic repair loops
• Full traceability with structured audit logs

---

Design Philosophy

Principle 1 — Information First

NeuroSym guards information, not prompts.
Inputs may originate from humans, tools, databases, or language models.

Principle 2 — Determinism by Default

Validation and repair operate fully offline.
Language models are invoked only when explicitly configured.

Principle 3 — Symbolic Core

Rules are explicit, testable, inspectable, and explainable.

Principle 4 — Auditability

Every decision produces a structured execution trace suitable for compliance, debugging, and research.

---

Installation

pip install neurosym-ai
pip install neurosym-ai[z3]          # SMT / formal constraints
pip install neurosym-ai[providers]   # Gemini / OpenAI adapters