Skip to content

Under active development

Development status →
TACITUS
Section pages

PRAXIS by TACITUS / Policy analysis workspace

A workspace for the whole policy assignment.

PRAXIS is an experimental workspace for research directors, analysts and editors. Review context from a document, attach it to a conversation and return to saved work. Shared team assignments and automatic source-change review are in development.

Try PRAXIS See the real workflowInspect a reference capsule

Experimental free access within app limits. Account required for saved work. Review AI answers. Paid upgrades are unavailable.

Real screens · synthetic private test material · September 30, 2026

From document to attached context.

Five chapters using three captured screens from a real-service test with synthetic material. Choose a chapter or move one step at a time. This is a guided sequence, not a recording or customer validation.

Chapter 1 of 5

Start with a source or purpose

An existing account created private context from synthetic TXT, PDF and DOCX documents or a purpose-only brief. Scanned OCR and long-document capacity were not established.

Real PRAXIS start with a source or purpose, September 30, 2026, build 4480650; synthetic acceptance-test material; tested upload, review, context attachment and saved-work scope
Open the full screen · No customer records or credentials are shown.
Read the complete tour transcript

Start with a source or purpose

An existing account created private context from synthetic TXT, PDF and DOCX documents or a purpose-only brief. Scanned OCR and long-document capacity were not established.

Inspect the proposal

Inspect proposed records and source references in the review drawer. Claims and graph changes remain proposals until reviewed.

Keep, reject or qualify

The test rejected a claim proposal and applied source receipts. Rebuilding unchanged inputs preserved decisions. Extraction accuracy still requires review.

Attach saved context and ask

A real conversation used the saved synthetic capsule. The answer recalled its project code and deadline, labeled as unverified user-provided context.

Return and review again

Saved conversations reloaded and reused context. Memo edit, save and reopen checks passed. Generated drafts required review; attribution and inference errors needed correction.

Acceptance source: PRAXIS release 4480650. These checks establish operation with the tested inputs. Read the scope and remaining gaps.

Example Advisory / synthetic local demonstration

A specialist distinction.
A better brief.

Brief a steering group by Day 16. Inspect the sources, apply a scoped method, review the proposed wording and carry the correction into the next task.

Runs deterministic browser checks on authored material. The real PRAXIS tour is separate. No model, live research, expert endorsement or customer result is implied.

Reporting-rule assignment · 32 declared members · seven neighboring dossiers excluded

Audience: Client steering group · Day 16 deadline · Day 15: July considered · no rule adopted

Read the passage before strengthening the claim.

Ministry of Reporting (fictional) / S01 §1

We propose a January 2027 start for the reporting rule. No rule has been adopted. An optional implementation pilot may be considered, subject to funding.
Source revision / original language
S01-authored-r1 / en / exact quotation, no translation
Event / publication / workspace capture
2026-10-01 / not separately supplied by the authored fixture / 2026-10-01
Applicable / valid interval
unknown; proposed dates are not adopted validity
Handling / rights
Authored synthetic passage; rights metadata does not grant redistribution permission.

Inspect this source in the complete record list →

Select a representation. All sources and actions in this workbench are local teaching material.

Follow C01 → E01 → S01 in the complete fixture

PRAXIS / the full workspace direction

Follow an assignment
through review and revision.

The director sets the question. An analyst compares the sources; a colleague corrects an inference. When a new source arrives, the earlier brief needs another look.

Concept preview · Product direction — interactive teaching screen, not a live app capture. The full team journey, change review and permissions remain development work.

Explore the eight-step workspace concept
PRAXIS / Example AdvisoryReporting-rule assignment · fictional

Research director / 1 of 8

A question, an audience and a deadline.

Example Advisory needs a decision-ready explanation for the client steering group by Day 16. The mandate is geopolitical-risk advice, not legal assurance.

PROJECT QUESTION

What should the client prepare for while the reporting rule is still under consultation?

Director → analyst → reviewing colleague → client steering group

S04 · assignment record

Compare with real tested screens →

One dossier / many useful outputs

Use the same dossier
for five different tasks.

Analysts, editors and directors use the same record for different work. These manually authored examples show what each deliverable must carry.

Inspect five authored deliverables

Fictional / authored output illustration / O02

Working notes

Audience
Analyst and editor
Purpose
Maintain a shared source ledger
As of
Day 15 · 15 October 2026 · fictional
Sources
S01 §1 · S02 §2 · S03 §1 · S04 §1–3

Day 1: January proposed [S01]. Day 8: industry requests transition [S02]. Day 15: July considered [S03]. Keep proposal, adoption and effective dates separate.

Unresolved / next review

Who will publish an adoption instrument, and when?

No rule adopted. Optional pilot funding undecided.

Illustrative revised text, not a PRAXIS-generated or approved output. The graph preserves the earlier Day 8 versions that need rechecking.

Current access and development boundaries

What you can evaluate

Current free access, subject to the limits shown in PRAXIS. Paid upgrades are unavailable. Future prices and features are proposals, not offers.

Inspect capability status and evidence
Upload documents, review capsule proposals, attach saved context, return to conversationsDocumented current core
Inspect source material before relying on an answer. Dated evidence →
Full team assignment and source-change review shown on this websiteConcept preview / product direction
The Example Advisory workspace and outputs are authored fictional illustrations. The complete project-linked team journey and automated source-change impact review remain development work; existing agent execution is a separate foundation.
User, Situation, Tool and Output capsule formatsReference formats
Four downloadable examples illustrate the model. They do not establish four complete PRAXIS authoring paths.
Conversational ontology editing and lossless interchangeIn development
Full round-trip portability needs validation.
Agent execution, queued research, approval and saved outputsExisting runtime / recorded bounded flow
Agent execution and durable run records exist. A queued research and owner-approval journey was recorded. General completion, source fidelity and the project-to-runtime bridge remain unvalidated. Dated evidence →
Team libraries, institutional controls and deployment optionsScope before commitment
Discuss permissions, data handling, support and feasibility before a pilot.
CONCORDIA and Wind TunnelExperimental
Synthetic modeled outputs do not establish live mediation performance, public opinion or behavioral validity.
Scoped briefing checks, saved reuse, terminology and method-pack review on this websiteLocal deterministic teaching demonstration
Authored browser checks and local JSON/storage lifecycle; no live model, production sharing, marketplace, credential verification or measured savings. Imported reviews require fresh local acceptance. Dated evidence →

Copy revised 2026-10-03. Evidence and verification scope · Versioned capability record

Further development

Carry the context into the next task.

We are developing richer graph editing, versioned context and handover between AI tools. Each step must preserve sources, qualifications and review state.

Read the design goals
Versioned context
Keep proposed changes, review decisions and earlier versions traceable.
Accountable review
Record who reviewed an artifact, its intended use and any caveats. Approval does not establish accuracy.
Reusable methods
Carry an analyst’s checklist or a mediator’s method alongside the source material.
Shared libraries
Preserve attribution, access rights and review state when context moves between people or tools.
DIALECTICA engine designRegistry proposalAnalytical methods

Project-linked context bridge · in development

Give an agent a task. Keep the context and the judgment.

PRAXIS already has bounded agent execution, durable runs and approval paths. The next bridge must pin a task packet from a capsule to the existing runtime. The proposed project workflow returns an artifact with exact source revisions for review. These examples explain the intended workflow; they do not run agents or generate findings.

Illustrative assignment

Where do the accounts of this event diverge?

  1. 01

    Attach the capsule

    The original statements, reporting dates, attributed claims, and an analyst’s notes on disputed evidence.

  2. 02

    Set the agent’s task

    Compare the accounts. Identify differences in chronology, attribution, and framing; retain the source passage for each.

  3. 03

    Review the proposal

    Check that a difference in wording has not been mistaken for a contradiction. Leave intent and coordination unresolved without supporting evidence.

  4. 04

    Keep the work record

    A comparison table with source references, review notes, and questions that need further investigation.

A reviewed result could inform the next task. Updating the capsule would remain an explicit review decision. The project bridge, complete source fidelity and general analytical task reliability still need end-to-end validation.

Inspect a reference Context Capsule

Existing PRAXIS runtime / unvalidated adapter

From a pinned capsule to the existing run queue.

PRAXIS already has an executing agent worker, durable queue/run records, approvals and saved output foundations. The missing bridge must select task context from an owner capsule, bind the returned artifact to its source revisions, and require human review of that exact revision.

Pinned owner capsule → selected task packet → existing queue/run request → bounded runtime → proposed artifact → exact-revision review → saved output with dependencies.

Inspect current states and future acceptance

Repository inspection: source 8ff509b8, 3 October 2026. Components: agent-runtime/worker.ts, cloud-tasks.ts, postgres.ts and types.ts. Run states: queued, running, waiting_approval, succeeded, failed, cancelled, expired. UI dispatch/ledger failure states are separate from this stored status union.

Canonical owner envelope schema 3 and older capsule/v2 compatibility are distinct from tacitus-web-context/1, the additive website demonstration. No production adapter, public API contract or runtime mutation is implemented here.

Future acceptance: resolve pinned source versions; check owner access and rights; reject stale review; handle unavailable sources, cancellation, failure and duplicate completion; preserve idempotency and durable output dependencies. Repository presence and dated synthetic operations do not prove general analytical quality.

Read the narrow adapter acceptance contract →Inspect dated operational evidence →
Try PRAXIS Discuss a team evaluationCheck data handling

Frequently asked

PRAXIS · frequently asked

The core workflow, intended users, and features still in development.

What is PRAXIS?
PRAXIS is an agentic workspace under development, built around Context Capsules for narrative analysis, statecraft, and cognitive analysis. The current workflow centers on creating and reviewing capsule context and using it in conversations. The agentic direction adds bounded tasks, returned artifacts, and human review.
Why organize work around a Context Capsule?
A capsule keeps relevant source material and reviewed context together so you can reuse it across questions. The broader design adds structured relationships, methods, and review history. Answers still need to be checked against the sources.
Who is PRAXIS intended for?
We are building for policy analysts, mission teams, mediators, business-risk teams, legal professionals, and researchers. These are intended use cases, not a list of confirmed customers. Contact us to discuss whether your workflow is a useful fit.
What does agentic mean in PRAXIS?
Our intended workflow lets a person give an agent a specific task and the Context Capsule it needs. The agent would return a proposed result with its supporting record. A person reviews the result and decides whether the capsule should change. Reliable delegation and this complete review loop remain in development.
Which capabilities are still being developed?
Semantic authoring, reliable delegated agents, complete citation-preserving exports, and shared team libraries remain development work. The availability section on this page distinguishes the core workflow from broader plans. A specific use case needs its own evaluation.
Is PRAXIS open source?
PRAXIS is a commercial product under development. TACITUS also publishes reference artifacts and research code. The open-source page identifies those projects; each repository defines its own license and scope.
Search and terminologyMachine-readable index

Optional analytics help us understand visits and product links. They load only with your permission. You can browse with them disabled and change your choice here.

Read privacy notice