Under active development · Building PRAXIS and the knowledge behind it.
Development status →Loading TACITUS
TACITUS / Context for agentic analysis
Decisions depend on more than an answer. They depend on the evidence, the people, and the history behind it.
We are building PRAXIS, an agentic workspace organized around Context Capsules: reviewable packages of sources, relationships, and methods. For statecraft, policy, and narrative analysis—and the decisions that depend on them.
Under active developmentSee current capabilities
case-nordalia-2026-03Diplomatic cable · ceasefireSource · case-nordalia-2026-03
¶1 / 1Typed graph · 0/8 primitives
0%Primitives extracted
Contradictions surfaced
scanning…
Episodes · illustrative score
composing…
Illustrative extraction using synthetic scenarios. Explore the idea behind Context Capsules: evidence, relationships, and reasoning kept together.
Evidence you can inspect
Keep the source close to the claim.
Disagreement you can examine
Compare accounts without erasing uncertainty.
Context you can build on
Carry reviewed knowledge into the next task.
What we are building
A new brief often starts with familiar questions: where did this claim come from, what changed, and why did we reach that conclusion? We are building a workspace that keeps those answers close to the work, and a knowledge layer that helps people carry reviewed context into the next task.
Where you bring documents, ask questions, review context, and return to saved conversations.
The aim is to assign agents bounded tasks, with a capsule supplying the sources, methods, and review requirements. Reliable delegated completion remains development work.
A saved package of context. The evolving design connects the material an analyst needs to inspect:
Sources
The documents and passages behind a claim.
Relationships
People, events, and competing accounts.
Methods
Questions, assumptions, and ways to reason.
Review
What was accepted, disputed, or left open.
Research and development on turning sources into structured proposals, preserving evidence, and recording review decisions. Integration and portability are still evolving.
This map explains the intended relationship between the projects. It does not imply a completed end-to-end integration. See documented capabilities and open work.
Built around the work
Our starting point is policy and conflict analysis. The same need runs through business, law, research, and engineering: understanding a situation without losing the evidence behind it.
Government · diplomacy · public affairs
These are intended applications, not customer deployments. Domain-specific performance will be evaluated with practitioners.
One structure · four types
Four ways to organize the context a task needs. Download the illustrative JSON files to inspect the proposed format. Support for the full design is still in development.
Who you work as: role, mandate, audience, voice, standing positions.
Built from: A conversation plus your past documents.
example .jsonWhat is going on: actors, claims, events, contradictions, episodes.
Built from: Your documents, feeds, and a guided interview.
example .jsonHow an expert thinks: methods, heuristics, traps, checklists, devices.
Built from: An expert-elicitation interview.
example .jsonWhat good looks like: format, structure, citation style, register.
Built from: Exemplar documents and a contract.
example .jsonThe proposed anatomy · nine layers
Fictional case · proposed workflow
Follow a fictional cable through five illustrated stages. The records and counts are prepared examples, not live extraction results. The full integration is under development.
Example records
Act 1 · A cable landsActors
0
Claims
0
Events
0
Commitments
0
Episodes
0
Allen relations
0
Contradictions
0
Review decisions
0
Citations
0
Devices applied
0
Inside a capsule · design in development
The capsule design separates source material, claims, relationships, and time. It also makes room for an expert’s methods and review decisions, so the next person can understand how an assessment was reached.
The goal is to make context reusable without losing its qualifications: who said what, what remains disputed, and what changed. Those records support review; they do not guarantee that an AI answer is correct.
Open formats · integration in development
We publish reference files so developers can examine the design. Planned integrations with tools such as MCP need to preserve source references, review decisions, and time records. Compatibility must be tested for each workflow.
TCGC is our benchmark project for reasoning with structured knowledge. Its public repository describes the approach. We have not yet published baseline results; performance remains to be established.
status: benchmark in progress — methodology published, baseline runs pending
Ceasefire Verification Heuristics
Tool capsule concept · verification questions · Concept example · not yet available
Regulatory Assumption Checklist
Tool capsule concept · assumptions to examine · Concept example · not yet available
Decision Memo House Style
Output capsule concept · citation requirements · Concept example · not yet available
We are exploring a curated collection of reusable methods with clear authorship, versions, and review history. These examples show the idea; the registry is not yet open.
Built for serious knowledge work
Options memos and assessments with sources a decision-maker can check.
A record of commitments, disputed accounts, and the conditions attached to an agreement.
A way to distinguish when events happened from when the evidence became available.
Legal, risk, regulatory affairs, investigations — wherever “source?” is the first question.
Why us
TACITUS starts from a practical problem: much of an expert’s judgment stays in their head. We are building ways to record the sources, assumptions, and methods behind an assessment, so others can examine and build on it.
ONTOLOGY PLAYGROUND · ILLUSTRATIVE EXAMPLES
Choose a fictional example and click Structure it to explore its actors, claims, and commitments. These annotations are prepared examples of the DIALECTICA design, not a live extraction service.
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 international 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
The highlights show where each annotation comes from. The counts are part of the prepared example, not a measure of extraction accuracy.
The intended workflow moves from source text to structured context, then to a brief. An analyst needs to review the interpretation at each step. This integration is in development.
Research in context / reviewed 29 September 2026
Recent work in information integrity and AI evaluation sharpens the questions we need to answer. These sources inform our research on narratives, statecraft, and reviewable AI assistance. The evaluation questions below are our own.
NIST · ARIA / 18 SEPTEMBER 2026
The ARIA Evaluation Planning Manual combines model testing, red teaming, and user testing. It starts by defining the sector, use case, and specific characteristic being measured.
Our evaluation questionCan an analyst using the workspace find errors, revise a judgment, and hand over the evidence reliably?
Read the source : Evaluate the application in the work it will do.NIST · EVALUATION PROBES / UPDATED 5 MAY 2026
NIST’s experimental probes examine faithfulness, completeness, and sufficiency: whether a source supports a claim, whether important context is missing, and whether the evidence is strong enough for the conclusion.
Our evaluation questionCan we catch a real quotation being used to support an inference it cannot justify?
Read the source : A citation has more than one job.EEAS · FOURTH FIMI REPORT / 12 MARCH 2026
The report examines information-manipulation cases documented during 2025. Its definitions distinguish incidents, campaigns, infrastructure, and activity sets from the actors who may be responsible.
Our evaluation questionCan a brief show observed source relationships without presenting uncertain coordination or attribution as fact?
Read the source : Trace an activity before attributing an actor.FRANCE · SGDSN / STRATEGY · 2026–2030
France’s strategy connects information resilience, education and research with the detection and characterization of foreign interference. It frames the response around a free, open information space.
Our evaluation questionCan analytical tools make evidence and uncertainty easier to examine while preserving independent judgment?
Read the source : Resilience includes the capacity to judge.Independent publications and public initiatives. The SGDSN link is its English courtesy translation. No affiliation, endorsement, or validation of TACITUS is implied.
Essays on how analysts work, how disagreements can be represented, and what an AI system needs to keep track of time and evidence. The notes explain the design choices and open questions behind AGON, KAIROS, and DIALECTICA.
Browse Notes by TACITUS
Why checking an answer requires more than a citation. This research note explores how source records, review decisions, and explicit uncertainty could make context easier to inspect.
What a source said, what an analyst concluded, when each became known, and which method produced the judgment are different kinds of statement. Why capsules keep nine named graphs with distinct mutability contracts instead of one big graph.
Can interviews and examples preserve useful parts of professional judgment? A research proposal for documenting methods, attributing them, and testing their limits in a different setting.
WORK WITH TACITUS
Start with your documents, attach saved context, and inspect the answer. Read the product guide or discuss a focused team pilot. PRAXIS remains under active development.
OPEN THE FLAGSHIP
Upload source documents, review capsule proposals, attach context to chat, and return to your saved work.
Open PRAXISTALK TO THE FOUNDER
For institutional pilots, domain datasets, mediation workflows, integration needs, or product partnerships, write directly to hello@tacitus.me.
hello@tacitus.meREAD THE IDEAS
The editorial home for the thinking behind TACITUS: conflict, structure, policy, diplomacy, and the grammar of disagreement.
Visit Concordia DiscorsPRAXIS and the supporting research projects are under active development. Explore the current capabilities and share what your work needs.
Read the open source