Loading TACITUS
PRAXIS · the flagship · in development
The AI analyst workbench for statecraft, policy, and mediation.
We build the context layer for serious policy and conflict work. PRAXIS is the workbench. It runs on DIALECTICA, a neurosymbolic engine composed of two open Rust layers — AGON for conflict vision and KAIROS for temporal vision — bound by an open ontology.
No black box. Every actor, claim, commitment, and event traces to a source — so an analyst, a mediator, or a desk officer can argue with the model rather than be argued at by it.
Cited at the span level.
Every claim, commitment, and event traces to source text — paragraph, line, timestamp.
doc-042 · ¶3 · 2026-03-12Time-ordered, causally linked.
Allen-13 temporal relations between every pair of episodes; causal chains kept explicit.
episode-014 · OVERLAPS episode-015Statecraft and mediation built in.
The kernel ontology is the trained reasoning of practitioners — not generic NLP entity types.
ACO v2.0 · 8 primitives · 41 classesDIALECTICA · live extraction
case-nordalia-2026-03Diplomatic cable · ceasefireSource · case-nordalia-2026-03
¶1 / 1Typed graph · 0/8 primitives
0%Nordalia
Suspend
Violation
DRM
Denial
Dialogue
Detainees
Primitives extracted
Contradictions surfaced
scanning…
Episodes · confidence
composing…
Startup program partner:Google Cloud for StartupsNeo4j Startup ProgramDatabricks for StartupsPRAXIS, DIALECTICA, and side projects under active development.
DIALECTICA · build#3171 · 25m ago
DIALECTICA · THE NEUROSYMBOLIC ENGINE
The credibility layer underneath PRAXIS.
DIALECTICA is the engine that makes PRAXIS reason over conflict the way the work demands — typed, time-ordered, source-bound. It is the structural alternative to fluent prose with no provenance.
Inputs
Reports
policy briefs, field notes, diplomatic cables
Dialogue
transcripts, mediation sessions, meetings
Evidence
claims, citations, datasets
Signals
media, narratives, polarization markers
Extraction target
Convert unstructured policy and conflict material into source-bound graph context before any LLM writes a sentence.
Dialectica · context graph
The LLM receives graph-shaped context, not a wall of prose.
Actor A asserted Claim B after Event C, constrained by Rule D, with source spans attached. The model can reason with structure already built.
Tools on top
PRAXIS · Flagship
AI analyst workbench for statecraft, policy, mediation
Wind Tunnel · Side project
Synthetic publics. Reception and polarization risk.
CONCORDIA · Side project
Live deliberation intelligence for the room.
Every tool can hand the same graph back to Claude, ChatGPT, Gemini, or an internal agent as structured context.
GND
Grounded facts
CTX
Context graph
EVD
Evidence & provenance
RZN
Reasoning layer
Scroll · the case file
From a single cable to a cited memo.
Watch one diplomatic cable move through the full TACITUS engine. Each act builds on the last, and the analyst dashboard on the left shows exactly what was extracted at each step.
Act 1
Raw source
A cable lands
A diplomatic cable arrives in the inbox: 9 paragraphs, 612 words, 4 referenced annexes. No structure, no provenance markers, no temporal order.
Act 2
Extract
AGON · conflict vision
AGON parses the prose. It marks actors, claims, denials, commitments, escalation signals — every typed primitive bound to its source span. Disagreement is preserved as data, not smoothed into summary.
Act 3
Order in time
KAIROS · temporal vision
KAIROS arranges the typed events on a dated axis. It computes Allen-13 relations between every pair of episodes — precedes, overlaps, meets, contains. The timeline is now machine-legible.
Act 4
Compose graph
DIALECTICA · neurosymbolic engine
DIALECTICA binds the conflict and temporal layers against the ACO kernel grammar. Contradiction edges are drawn explicitly. The graph carries four reasoning layers — ground, context, evidence, reasoning.
Act 5
Cited memo
PRAXIS · analyst workbench
PRAXIS drafts a SITREP on top of the graph. Every paragraph carries source citations. Every assumption is editable. Every open question is visible. The model serves the analyst — not the other way round.
Extraction telemetry
Act 1 · A cable landsActors
0
Claims
0
Events
0
Commitments
0
Episodes
0
Allen relations
0
Contradictions
0
Citations
0
cable · classified · draftingdoc-042 · 12 mar 2026 · 1 / 1
On 12 March 2026, the Government of Nordalia issued a démarche
through its Ministry of Foreign Affairs suspending negotiations with the
Democratic Reform Movement (DRM) and citing repeated ceasefire violations near
the 4.2-km demarcation line. The DRM rejected the breach allegation the same day,
characterising the movement as administrative reposition under §4 of the Framework.
Both parties confirmed their commitment to continued dialogue, with the DRM linking
progress to release of detained political negotiators on or before 18 March.
612 words · 4 annexes · no structure
AGON · extracting typed primitivesdoc-042 · spans bound
On 12 March 2026, the Government of Nordalia issued a démarche
through its Ministry of Foreign Affairs suspending negotiations with the
Democratic Reform Movement (DRM) and citing repeated ceasefire violations near
the 4.2-km demarcation line. The DRM rejected the breach allegation the same day,
characterising the movement as administrative reposition under §4 of the Framework.
Both parties confirmed their commitment to continued dialogue, with the DRM linking
progress to release of detained political negotiators on or before 18 March.
ActorClaimDateEventCommitmentConstraint
KAIROS · temporal vision · 7 dated mentions · 6 episodes · 22 Allen relations
timeline ▸ 11–18 MarDIALECTICA · composed graph · 7 nodes · 7 typed edges
3 contradictions · layered: GND · CTX · EVD · RZNGND
CTX
EVD
RZN
Nordalia
Actor
Suspend
Event
Violation
Claim
DRM
Actor
Denial
Event
Dialogue
Commitment
Detainees
Interest
PRAXIS · SITREP draft · 11 citations · all spans bound
house-style · UN-citation · provenance preservedSITREP — Nordalia / DRM · 15 Mar 2026 · TS-LIMDIS
Headline.Nordalia suspended bilateral negotiations on 12 Mar[doc-042 ¶2] citing alleged ceasefire violations near the 4.2-km demarcation line[doc-042 ¶3]. The DRM rejected the breach allegation the same day[doc-044 ¶1] and characterised observed movement as administrative reposition under §4 of the Framework[doc-044 ¶3, art-04].
Contradictions surfaced.The model registers a direct contradiction between the ‘violation’ claim and the DRM denial within the same 24-hour window[edge-c1↔e2]. Verification scope remains contested under Article 4.
Commitments tracked. Both parties confirmed continued dialogue[doc-042 ¶5; doc-044 ¶4]. The DRM conditioned progress on release of detained political negotiators on or before 18 Mar[doc-044 ¶5].
Open. Verification mechanism for the 5-km halt unspecified; reciprocity clause requires Article 4 cross-reference.
Act 1 · A cable lands
case-nordalia-2026-03 · doc-042AI FOR STATECRAFT · THE INVESTABLE THESIS
The next policy platform is a knowledge graph with an analyst on top.
TACITUS is building graph-native AI infrastructure for teams whose mistakes have institutional consequences. The wager is concrete: if statecraft work depends on evidence, sequence, incentives, and commitments, the AI layer must preserve those structures before it writes.
The statecraft problem is structural.
Diplomatic, policy, and mediation teams do not just need summaries. They need a shared model of actors, claims, interests, commitments, events, constraints, leverage, and narrative drift.
Grounding must happen before generation.
Flat retrieval can find text. TACITUS first builds a provenance-bound graph, then lets models reason against relationships that already exist.
Human judgment stays in charge.
PRAXIS is decision support, not automated diplomacy. Every output is designed to be inspected, contested, edited, exported, and traced back to source.
Research spine
Built as an experimental product, argued as a serious infrastructure company.
The case for TACITUS is not “chat for diplomats.” It is a governed context layer for policy work: open conflict assessment practices, graph reasoning, grounded retrieval, risk-managed AI, and human-led mediation intelligence brought into one workflow.
CSIS · AI and the Future of Mediation
AI can support mediation when it preserves confidentiality, neutrality, and human control.
Microsoft Research · GraphRAG
Graph-based indexing improves corpus-level sensemaking beyond isolated document retrieval.
ICLR 2024 · Think-on-Graph
Knowledge-graph reasoning paths make LLM answers more explainable and traceable.
USAID · Conflict Assessment Framework
Conflict assessment starts with actors, grievances, incentives, institutions, and dynamics.
NIST · AI Risk Management Framework
High-stakes AI should be governed, mapped, measured, managed, and auditable.
UN-World Bank · Pathways for Peace
Prevention requires connecting political, social, economic, and security drivers early.
ONE FLAGSHIP · ONE ENGINE · TWO SIDE PROJECTS
PRAXIS is the workbench. DIALECTICA is the engine underneath.
TACITUS is graph-native AI infrastructure for statecraft, conflict, and policy work. PRAXIS is the analyst-facing workspace. DIALECTICA is the trust graph that makes it reason instead of guess. Wind Tunnel and CONCORDIA apply the same backbone to public reaction risk and live deliberation.
SIDE PROJECTS · SAME BACKBONE
Side projects, same backbone.
Wind Tunnel and CONCORDIA are focused experiments built from the same backbone: reaction-risk modeling before a public move, and live deliberation intelligence while a process is underway.
PRAXIS, DIALECTICA, Wind Tunnel, and CONCORDIA are all under active development. Send feedback.
ONTOLOGY PLAYGROUND · INTERACTIVE
Paste a policy crisis. See the structure.
Pick a statecraft sample, click Structure it, and watch DIALECTICA tag the primitives PRAXIS needs: actors, claims, interests, commitments, constraints, leverage, events, and narratives, with source-span grounding.
INPUT · policy text
Ceasefire allegations, humanitarian access, guarantor pressure.
The Ministry of Security alleged on 18 April that the Northern Coalition violated the ceasefire near Kalo Bridge. The coalition denied responsibility and reaffirmed its written commitment to keep the southern corridor open. The UN mission warned that aid access depends on a 72-hour notification window, while both parties need credible verification before the next donor conference. The government is framing the incident as proof the coalition cannot be trusted.
8 PRIMITIVES
GRAPH STATS · typed output
-
Actors
-
Claims
-
Commitments
-
Contradictions
Every node is bound to the source span it came from. Every edge is typed. Every claim is auditable back to the input.
Raw memo
ACO graph
PRAXIS brief
Extraction is not the final answer. It is the grounding layer: raw policy text becomes a typed graph; the graph becomes a brief the analyst can inspect, edit, and challenge.
NOTES BY TACITUS
Working notes on the engine, the ontology, and the work itself.
Short, technical, citation-bearing essays. AGON and KAIROS internals. How mediators and policy analysts think. Bi-temporal honesty. Why generic LLMs break on conflict. Every note has a TL;DR, sources, and a specific problem it answers.
Browse Notes by TACITUS
2026-05-13Positioning
Why TACITUS — and why the LLM-only stack is structurally insufficient.
Policy and political work is knowledge work over typed structure. The dominant 2024–2026 stack fails on five measurable axes — time, causality, provenance, contradiction, long-horizon context. A larger model patches none of them. Here is what the corrective looks like.
2026-05-13Engine
AGON — the conflict-vision crate inside DIALECTICA.
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.
2026-05-13Engine
KAIROS — temporal vision and the discipline of two clocks.
KAIROS implements Allen-13 interval logic, bi-temporal commitment status, and as-of / drift / retrospective queries. It is what makes the LLM-side claim that something happened on Thursday survive contact with reality.
WORK WITH TACITUS
Pilots, demos, partnerships, and technical integrations.
TACITUS is building with policy teams, mediators, researchers, builders, and institutions that need better context infrastructure for complex situations.
OPEN THE FLAGSHIP
PRAXIS — analyst workbench.
Statecraft, policy, mediation. Cited briefings. Conflict graph. Structured reasoning you can argue with — not be argued at.
Open PRAXISTALK TO THE FOUNDER
Discuss a pilot or walkthrough.
For institutional pilots, domain datasets, mediation workflows, API access, or product partnerships, write directly to hello@tacitus.me.
hello@tacitus.meREAD THE IDEAS
Concordia Discors.
The editorial home for the thinking behind TACITUS: conflict, structure, policy, diplomacy, and the grammar of disagreement.
Visit Concordia DiscorsPRAXIS, DIALECTICA, and the side projects are under active development. Expect edges. Send feedback.
Read the open source