Conflict and policy intelligence infrastructure should be transparent and extensible. Three public repositories cover the pipeline, the engine reference, and the ontology spec. Read them, run them, critique them, fork them.
Open stack
TACITUS can be a company and still make the core grammar inspectable. The open-source work exists so researchers and practitioners can see how policy and conflict context becomes a graph before a product uses it.
Spec
The shared grammar: primitives, subclasses, relationship types, graph layers, and serialization rules.
Pipeline
Source ingestion, primitive extraction, confidence handling, source-span binding, and schema validation.
Graph
Temporal graph construction, contradiction edges, commitment state, graph queries, and context packs.
Products
PRAXIS is the flagship workbench. Wind Tunnel and CONCORDIA are side projects. All run on the same DIALECTICA backbone.
Repositories
From a simplified knowledge pipeline to the full neurosymbolic engine. Pick the level of depth that fits your use case.
sargonxg/TACITUS-Knowledge-Pipeline-open
The reference extraction and structuring pipeline: ingest, extract, structure, query, implemented end-to-end with the eight-primitive ontology and sample conflict datasets. The fastest way to run TACITUS locally.
What's inside
sargonxg/A2_DIALECTICAbyTACITUS
Reference implementation of the Dialectica engine, the neurosymbolic core that powers every TACITUS product. Ontology Augmented Generation (OAG), typed knowledge graph construction, multi-hop reasoning, full provenance.
What's inside
sargonxg/tacitus-ontology
The Agentic Conflict Ontology as a standalone spec. Pydantic schemas, OWL/Turtle dual export, RFC process for revisions. The open grammar every TACITUS product shares.
What's inside
Architecture
All three repositories share the same foundation: a typed, provenance-bound graph grounded by the Agentic Conflict Ontology, augmented by language models for extraction and fluent output.
8 primitives, 41+ typed classes. The schema that makes conflict machine-legible.
GND (ground truth), CTX (context), EVD (evidence), RZN (reasoning). Each layer serves a different intelligence function.
The graph never hallucinates. The LLM never flies blind. Deterministic facts plus probabilistic inference.
Every assertion traces back to a source document, timestamp, and actor. Audit trail by design.
We are a small team building something that does not yet exist: AI infrastructure purpose-built for conflict and human-friction intelligence. Engineers, conflict professionals, researchers: if the brief sounds right, we want to hear from you.
Fork a repo, open an issue, submit a PR, or write in with ideas. The best conflict ontology will be built by the community that uses it.
Engineers
Python, TypeScript, Neo4j, LLMs, graph databases, neurosymbolic AI.
Conflict professionals
Mediators, negotiators, diplomats, HR; domain expertise matters.
Researchers
Ontology engineering, conflict theory, computational social science, NLP.
Ideas and contributions welcome