We publish the core so the claims can be run, not just cited. The symbolic reasoning engine — the part where "0% hallucination on the provable path" has to be true — is open under the MIT license.
nusy-reasoners
Proof-carrying reasoning over Apache Arrow. Derivations you can audit; abstention you can trust. When the symbolic path can't prove an answer it declines rather than fabricates — so on that path, hallucination is zero by construction, not by hope. This is the symbolic core of the NuSy reasoning model.
Rust · MIT License.
OpenDrugGraph
A community drug-interaction graph with never-launder provenance. 513 drug-interaction edges derived from FDA Structured Product Labels — every edge cited to its exact label source span (SET_ID + LOINC section + sentence), or absent. Mechanism-level edges (CYP/transporter) are first-class, and the extraction never fabricates a “safe” answer: unconfirmed or un-citable interactions are flagged, never asserted. Transparency-first and early — coverage is growing.
Data + evidence · MIT (FDA/RxNorm source data is public domain).
Live app: opendruggraph.com · GitHub
nusy-graph
Document → Y-layer knowledge graph, as a product. The nusy-grapher library, CLI, and MCP server for turning documents into a queryable semantic knowledge graph. (MIT)
nusy-kanban
Arrow-native, distributed kanban for multi-agent teams. SHACL shapes, graph-native pull requests, dual boards, and NATS-backed single-writer coordination — git-as-database for a fleet of AI agents building in parallel. We use it to coordinate the fleet that builds NuSy, with single-writer semantics that prevent ID collisions across machines.
Rust · published on crates.io.
cargo install nusy-kanban
noesis-ship
A NATS-based multi-agent communication platform. A NATS core plus pluggable adapters (WebSocket, MCP, HTTP/SSE) — the generic communication layer beings and agents talk over. NuSy uses it for every NATS-based interaction.
MIT License.
Foundations
The earlier, Python-era projects that seeded the approach — the knowledge format and the evaluation framework that the Rust/Arrow stack above grew out of:
- Yurtle — Markdown that becomes a knowledge graph: add a Turtle/RDF block to any Markdown file and it's a queryable node. Defines the Y-Layer specification (Y0–Y6). (MIT)
- yurtle-rdflib — an RDFlib plugin that parses Yurtle files into SPARQL-queryable graphs. (MIT)
- acf-framework — the AGI Certification Framework: a multi-dimension, measure-based diagnostic of AI understanding against human professional standards, with the Zorblaxia battery (fictional-domain tests that can't be gamed by training-data contamination). (MIT)
nusy-kanban(Rust, Arrow-native) supersedes the earlier Pythonyurtle-kanban.
How they fit together
noesis-ship The nervous system — how agents and beings talk (NATS)
+
nusy-kanban The workflow — who builds what, with graph-native review
+
nusy-reasoners The mind — proof-carrying reasoning you can audit
↓
NuSy beings The application — AI entities that learn, reason, and prove