Projects

Rust CLI tool with an MCP server built in. Indexes .sysml repositories into a persistent knowledge graph and exposes it through 14 CLI commands and 10 MCP tools. Single binary, dual mode: run it as a CLI for scripting and benchmarking, or as an MCP server for editor and agent integration.

9-signal hybrid index: keyword scoring (8 signals including exact match, prefix, containment, vocabulary expansion, relationship adjacency) plus fastembed all-MiniLM-L6-v2 vector search (384-dim, HNSW) and 27 SysML v2 structural relationship types. Achieves 94% average token reduction vs raw file injection.

Install

Key commands

Command	Description
sysml search	Hybrid keyword + vector search across indexed models
sysml trace	Follow structural relationships (specialization, usage, allocation)
sysml render	Pre-render element views for LLM consumption
sysml check	Completeness checking against SysML v2 structural rules
sysml inspect	Detailed element metadata and relationship graph

Evaluation harness measuring how CLI tool configurations affect LLM accuracy on structured systems engineering tasks. 132 tasks across 8 categories: discovery, reasoning, explanation, layer, boundary, vector-sensitive, structural trace, and corpus scaling.

Per-field structured scoring (Bool, Float, Str, ListStr F1 with threshold). Corpus: Eve Online Mining Frigate SysML v2 model, 19 files, 798 elements, 1,515 relationships. Scaling corpus: 95 files.

Install

Task categories

Category	Tasks	Tests
Discovery	Attribute lookup, element enumeration	Can the model find specific facts?
Reasoning	Multi-hop inference, constraint satisfaction	Can the model reason across relationships?
Explanation	Summarize structure, describe behavior	Can the model explain what a model element does?
Structural trace	Follow allocation, specialization chains	Can the model traverse the model graph?
Corpus scaling	Same tasks on 5× larger corpus	Does performance hold at scale?

Tree-sitter grammar for SysML v2. The parsing foundation for all Nomograph tooling. Built by curating a corpus of real-world SysML v2 models and iterating the grammar against it: run the parser, find failures, fix the grammar, repeat. 89% coverage on external files (models we did not author). 192 tests passing.

Provides incremental parsing with bindings for Rust, C, Node.js, Python, Go, and Swift. The sysml CLI and benchmark harness both depend on it. SysML v2 was adopted by OMG in June 2025; this grammar tracks the current specification.

Bindings

Language	Package	Status
Rust	tree-sitter-sysml	Primary (used by sysml CLI)
C	Header + source	Generated by tree-sitter
Node.js	tree-sitter-sysml	Available
Python	tree-sitter-sysml	Available
Go	tree-sitter-sysml	Available
Swift	TreeSitterSysml	Available

We are genuinely curious about the return to composable CLI tools for LLM interaction. There is something appealing about the Unix philosophy applied to AI tooling: small programs that do one thing well, piped together, with text as the universal interface. Our benchmark data suggests this intuition has substance. CLI tool-based search outperformed both MCP transport and RAG on discovery tasks, using 21% fewer tokens. The composable approach seems to produce less overhead and more predictable behavior.

The methodology is designed to extend to any formal language with a grammar. AADL for embedded systems, OSCAL for security compliance, KiCad for electronics, OpenSCAD for parametric 3D. The source code of the physical world, made legible to AI through the same tooling pattern: parse it, index it, expose it through composable CLI commands, measure what happens.