TL;DR
- 01AGON is the Rust crate inside DIALECTICA that does conflict vision: it reads a corpus, extracts typed primitives against the ACO kernel, and emits typed graph operations under validation.
- 02It is the half of the neurosymbolic stack that recognises *what kind of thing* a span of text is — Actor, Claim, Interest, Constraint, Leverage, Commitment, Event, Narrative — not what it means.
- 03AGON is paired with KAIROS (temporal vision). Together they cover the recognition surface; the typed graph below them carries the reasoning.
- 04Stack: Rust 1.79+, tokio, candle for in-process models, gRPC + protobuf service surface, MIT-licensed.
AGON is the half of the neurosymbolic stack that recognises what kind of thing a span of text is — actor, claim, interest, constraint, leverage, commitment, event, narrative — under a strict typed contract. Rust. MIT. Hot path.
What AGON is responsible for
- ▸Entity disambiguation in conflict contexts (alias resolution, role attribution, jurisdiction inference).
- ▸Speech-act classification: ASSERTED / DENIED / ACKNOWLEDGED / ACKNOWLEDGED_AMBIGUOUSLY / DENIES_SCOPE / COMMITS_TO / REVOKES.
- ▸Position-vs-Interest separation (Fisher/Ury distinction) at the type level.
- ▸Contradiction-pair surfacing across documents.
- ▸Leverage and constraint identification with mechanism + binding-strength fields populated.
What AGON is *not* responsible for
- ▸Time. That is KAIROS.
- ▸Validation against kernel invariants. That is the engine's validator.
- ▸Generation. AGON is recognition only. The Ontology-Augmented Generation pattern uses AGON as one input.
- ▸Editorial framing. AGON emits typed candidates with confidence scores; the analyst commits or rejects.
Why Rust
Three reasons. First, AGON runs hot — every document touched by the pipeline passes through it, and Python-bound model serving was the wrong floor for the throughput we needed. Second, the type system lets us encode the kernel invariants once and have the compiler check them across the codebase rather than at runtime. Third, the entire crate is statically linkable into customer deployments that cannot run Python — gov tenants, air-gapped envs, edge devices.
Open-source status
AGON is published under MIT alongside KAIROS, the TACITUS Knowledge Pipeline, the Agentic Conflict Ontology, and the TCGC benchmark. The reference engine binary is open. The case-specific tuning is not — partner data stays on-prem.
SOURCES
