Sourceability
Use PRAXIS to see which important claims have sources, which look weak, and which still need evidence before a brief is ready to share.
PRAXIS by TACITUS
How PRAXIS works
PRAXIS is the visible workspace where users ask policy questions, attach sources, save context, draft briefs, and keep the evidence trail visible. The surface stays simple: one place to turn a situation, document, or hard question into reviewable work.
The app first
From loose question to reviewable work
A user can begin with a prompt, a source packet, a draft question, or a document. PRAXIS keeps the sources, assumptions, gaps, and next actions near the answer, then produces a memo, brief, source review, or handover that still carries its evidence trail.
Use it today
Current PRAXIS stays simple: one cockpit, context attached, evidence visible.
The current app centers on Home, saved context, Ask PRAXIS, sources, memos, and receipts. Upstream TACITUS systems and deeper graph or temporal engines are described as integration layers or roadmap lanes when they are not the visible production surface.
Claim-source alignment
PRAXIS keeps source previews, gaps, and claim posture close to the answer so teams can separate cited facts from model judgment or user assumptions.
Role-aware context
Context Capsules can carry role, institution, lens, constraints, and output goals so PRAXIS answers for a real policy job instead of a generic chat.
Output quality
Brief, memo, source review, and handover workflows keep caveats, receipts, and next actions attached to the work product.
Thinking devices
ACH, timeline, causal, human-friction, red-team, boardroom, and sourceability devices can sit behind the simple Ask surface when the work needs them.
For technical readers
A Wikipedia for Agentic Context
ContextCapsules are reusable context objects, not prompts. They preserve the parts of policy work that a capable analyst would normally carry in their head: sources, assumptions, tacit knowledge, reasoning devices, review gates, and allowed agent behavior.
Agents can inherit a capsule safely because claims, sources, memories, and limitations are labeled. The capsule says what is known, what is only remembered, what is stale, what is disputed, and what an agent must stop and hand back for review.
Human review makes truth canonical
AI and Cognee can suggest improvements, recall advisory memory, find source gaps, or draft review cards, but human review decides what becomes canonical in the capsule.
Community capsules stay inspectable
Marketplace and community capsules should be forkable and inspectable, with provenance and review state visible before a team relies on them.
What the capsule carries
The capsule carries source spines, claim ledgers, tacit constraints, output contracts, allowed tools, forbidden actions, review checkpoints, memory receipts, and stale or rejected inferences as labeled context.
How generation works
PRAXIS-augmented generation: four capsules, one dock, agents that check their work
Every answer starts at the Capsule Dock: four typed slots above the Ask box. A User capsule says who is asking, a Situation capsule carries the case, a Tool capsule sets the allowed methods, and an Output capsule defines what the work product must look like. That is the whole user-facing model — fill the slots that matter and ask.
1 · The resolver compiles the dock
When you send a question, PRAXIS resolves each docked capsule — its sources, claim ledger, graph references, ontology binding, and review state — into one layered context block. The layers stay separate on purpose: the Situation capsule governs facts, the User capsule governs voice, the Tool capsule governs method, and the Output capsule governs format. Conflicts are surfaced, never silently blended.
2 · Retrieval is governed, not generic
The capsule decides what the model is allowed to read: uploaded documents, source receipts, the typed knowledge graph of actors, claims, and commitments, vector search over the situation corpus, and grounded web search when the capsule permits it. Outside model knowledge is labeled as supplemental — never presented as capsule fact.
3 · Agents extract, research, and check
Deep questions queue a capsule-bound agent run with a durable ledger: a research agent plans searches and gathers receipts, an extraction step proposes graph updates, a sourceability reviewer checks that claims have support, and a red-team pass challenges the draft. Risky steps stop at human approval gates, and every tool call is recorded so you can audit the run afterwards.
4 · Truth stays layered and reviewable
The answer separates capsule-grounded facts, retrieved evidence, model judgment, and your assumptions. Anything an agent learned — a new source, a graph claim, a corrected assumption — becomes a review card; only human approval makes it canonical in the capsule. That is why a docked capsule beats a plain LLM chat: delimited context in, receipts out, and the context itself improves with every reviewed run.
Premium web is not enabled by default.
PRAXIS can expose a paid source-discovery lane only when the deployment has a configured provider and the user explicitly starts that paid workflow. Until then, public claims stay limited to the wired source receipts, uploaded documents, grounded-search routes, and review gates available in the app.
GDELT + Wikidata add advisory entity and media signals.
Research graph is free in v1 for signed-in users. It grounds entities with Wikidata labels, aliases, identifiers, and official links, then checks GDELT DOC/API for media pulse and relationship hints. It never mutates the canonical PRAXIS graph automatically; candidates stay review-gated.
User
The analyst posture, preferences, role, institution, and working assumptions that shape the answer.
Situation
The live policy problem, actors, geography, constraints, stakes, and decision context.
Source
The evidence posture: what is cited, what is missing, what looks weak, and what needs review.
Document
Uploaded files, notes, excerpts, and source packets that should travel with the work.
Graph
Actors, claims, relationships, contradictions, commitments, and typed links that make the work reusable.
Time
Sequence, deadlines, windows, episodes, commitments, and signals that change the answer over time.
Tool
The policy tools, checks, drafts, and specialist routines that can run against the capsule.
Objective
The job to be done: explain, compare, decide, prepare, brief, challenge, or hand over.
Output
The memo, brief, talking points, scenario table, source review, or handover artifact the user needs.
Memory
The durable parts that should survive beyond one chat: decisions, gaps, corrections, and team context.
TACITUS ecosystem
PRAXIS is the workbench. The deeper systems organize what the workbench can use.
DIALECTICA, AGON, and KAIROS should be understood as upstream or intended integration layers where appropriate. PRAXIS is the public policy workspace that turns their structure into user-facing analysis and artifacts.
The broader conflict-intelligence company and ontology stack behind PRAXIS.
The visible policy workbench that turns structured context into questions, tools, memos, and handovers.
The ontology-augmented generation layer for ingestion, typed graphs, provenance, and deterministic reasoning.
An upstream perception and evidence engine, under development, for actors, claims, contradictions, commitments, friction, and quality gates.
An upstream temporal engine, under development, for event sequence, Allen-style relations, commitments, episodes, and time-aware signals.
The shared ontology work that helps keep products and research systems aligned around typed concepts.
Community
How users can contribute
The useful unit of contribution is not only code. Better sources, clearer capsule gaps, reviewed graph claims, and repeatable team context all make PRAXIS more useful.
Upload documents and source packets that make a situation easier to verify.
Correct missing capsule fields, weak source claims, or stale assumptions.
Review graph and source claims before they become reusable team context.
Create reusable capsules for recurring regions, actors, disputes, or institutional workflows.
Share product feedback from real policy, diplomacy, research, or briefing work.
Open issues or pull requests in the linked GitHub repositories when the contribution belongs upstream.
Aspirational
What I'd like PRAXIS to support next
The next version should make source-to-capsule saving one click, let teams share and fork capsules, review DIALECTICA graph deltas inside PRAXIS, show AGON and KAIROS evidence cards inside Ask answers, score capsule health, and support collaborative memo review. These are product goals, not a claim that every integration is already fully wired in production.