Nexxus
High-throughput semantic policy mapping and query routing.
Nexxus maps natural-language queries into structured AST constraints deterministically, serving as the bridge between prose policies and Anchor validations.
Overview
Nexxus resolves natural language policies into structured code patterns that Anchor can scan or intercept.
Problem
- Policies are written in human language, whereas security enforcement requires exact AST matching.
- Mapping policies to code manually creates gaps and translation errors.
Motivation
To make compliance rules accessible to policy makers without requiring them to write Tree-sitter query expressions.
Solution
A semantic router that classifies input goals and resolves them to validated AST match templates.
Implementation
Lightweight Python semantic classifier with low-latency caching wrappers.
Engineering Decisions
Caching mapping configurations to guarantee microsecond-scale execution inside request paths.
Challenges & Lessons
Validating that semantic translations don't introduce false negatives in rule matches.
Nexxus Translation
Active development
Natural Language Policy ↓ (Semantic Classifier) Policy Map Cache ↓ (Resolve Query) Structured AST Queries ↓ Anchor Code Validation
Resolves mappings deterministically.
Uses low-latency lookup tables.
Components
Functional building blocks
The flagship documentation uses the same component taxonomy across every system.
Classifier
Categorizes input goals.
Cache Router
Bridges queries with AST patterns.
Features
Core capabilities
These are product features in the engineering sense, not marketing copy.
Natural language policy mapping
✓Microsecond cache routing
✓Routing Check
Map a rule query.
import nexxus
nexxus.map_policy("prevent raw completions API calls")Example output
A governed action should return enough evidence to review without reconstructing the entire runtime path.
Resolved category: API_SECURITY AST Query: (call_expression) @completions
Documentation
- Nexxus API guide
Papers & Consultation Briefs
Benchmarks
- Cache lookup: <20µs
- Classification: <15ms
Roadmap
The roadmap separates what is already shipped from the next layer of work so the public record stays explicit.
Shipped
Core mapping logic · Low-latency cache
Next
Direct Canon schema compilation · Visual query editor for mapping rules