Loading TACITUS
Loading TACITUS
Ask normally. Attach saved context — Context Capsules carrying your sources, your situation, and your experts’ reasoning. Leave with reviewable, cited work: briefs with receipts, review gates on by design, and context your team keeps.
Powered by DIALECTICA, the open-source capsule engine. Structured. Auditable. Contestable.
Where PRAXIS is going
Wikipedia proved that versioned, cited, openly-reviewed knowledge can outcompete closed encyclopedias. But Wikipedia articles are prose about settled facts. The knowledge institutions actually run on — live situations, expert methods, house standards — has no equivalent. PRAXIS is being built to be that equivalent: a workspace where every unit of knowledge is a Context Capsule — versioned like an article, cited like a good one, reviewed by named experts instead of anonymous editors, and attachable to any AI workflow instead of only readable by humans.
Today that means your team’s private library: fork a starter capsule, build your own from documents and interviews, review what the engine proposes, and reuse across every ask. Tomorrow it means organization-wide libraries with lineage and attribution — and eventually the open Capsule Registry, where a mediator’s heuristics or a regulator’s checklist can be published, cited, and built upon like any other reference work.
Versioned
Capsules carry their history: what was proposed, what a reviewer approved or rejected, what superseded what. Knowledge with a changelog, not a final draft.
Reviewed, with names
Trust tiers and review gates replace anonymous edit wars. You can see who vouched for a claim, how strongly, and with what caveats.
Readable by machines
An encyclopedia humans read; a capsule library both humans and agents use — typed structure, embedded graphs, and a runtime contract on every object.
A diplomatic file, a peace process, a policy controversy — the structure already exists in someone’s head. PRAXIS is where it stops living only there.
STATECRAFT WORKFLOW
PRAXIS is designed for the moment when a desk officer, mediator, or policy team has too many documents and too little shared structure. The model does not become the authority. The graph becomes the inspection surface.
01
User Core (role, audience, conventions) + Situation Core (the active file). PRAXIS routes from there.
02
Sources, episodes, actors, claims, evidence — ingested, typed, time-ordered, provenance-bound.
03
Ask PRAXIS for cited answers. Run ACH on competing hypotheses. Promote Boardroom synthesis only after review.
04
Country cards, options memos, SITREPs, stakeholder maps, commitment ledgers, briefing packs — exported with provenance.
What changes
A briefing is no longer a model summary. It is a view over a cited conflict graph: every paragraph can point back to evidence, every assumption can be contested, and every commitment can be checked later.
CAPABILITIES
Ask PRAXIS · Cores · Context Workbench · Graph Intelligence · Memo Studio · Boardroom · ACH · Exports. The shipped app, not a chat box.
The authenticated first screen. Cited answers, open questions, next-step routing — all grounded in the active User Core and Situation Core. Not a chatbot. The starting point of the workday.
Attach the context that actually matters: who you are (FP analyst, policy lead, mediator) and what file is active (Indo-Pacific competition, AI regulation, peace process). PRAXIS routes the next move accordingly.
Sources, claims, evidence, actors, episodes, and drafts stay visibly connected. The graph is not decoration — it is the routing layer for what gets researched, challenged, drafted, and reviewed.
Turn findings into reviewable memos with versions, exports, evidence, and gaps visible. House-style templates. Provenance survives every export.
Frame, sources, lenses, work process, timeline, graph, drafts, and methods in one shell. The case shell that holds a file from intake through drafting.
Run advisor deliberation. Multiple analyst agents argue the file; you promote synthesis only after human review. Not autonomy — supervised disagreement.
Structured ACH (Analysis of Competing Hypotheses) inside the workbench. Stress-test hypotheses against evidence that supports and weakens each explanation. A trained method, not a vibe.
No black-box conclusions. Every assertion is editable, traceable, refutable. The model serves the analyst — not the other way round.
Hover the highlighted spans. Each is a typed primitive in the conflict graph — actor, claim, commitment, interest, event — bound to its source.
The Government of NordaliaActor announced that it would suspend all negotiationsEvent with the Democratic Reform MovementActor, citing repeated violations of the ceasefire agreementClaim. The DRM denied the allegationsClaim and stated their commitment to dialogueCommitment remains conditional on the release of political prisonersInterest.
Built for desks, missions, and mediation teams. Under active development. Send feedback.
Frequently asked
Common questions from policy desks, mission teams, and mediators about the analyst workbench.