How PRAXIS works

Ask a policy question, inspect its evidence, then save the background you want to reuse.

What is a Context Capsule?

A capsule is saved context you can attach to chat. It holds background, sources and review notes, so you can reuse them without explaining everything again. You can also ask without one.

Purpose: The question, audience and output you need.
Evidence: Source references, claims and supporting excerpts.
Connections: Recorded actors, relationships and events.
Review: Decisions, assumptions and unresolved gaps.

Start from notes, a template or TXT, text-based PDF and Word files. Review the proposed facts, save the capsule, then choose Use in chat.

Create a capsule

One workspace for the question, context and evidence.

Use it today: Chat, Capsules and saved work. Sources can be incomplete or wrong; check important claims against the original material.

Sourceability: Check which claims have support and which still need evidence.
Claim-source alignment: Read claims alongside source previews and recorded gaps.
Role-aware context: Save a role, audience, constraints and analytical approach for reuse.
Output quality: Edit and review saved work before sharing. Structure checks do not measure factual accuracy.
Thinking devices: Choose a method when useful: compare hypotheses, build a timeline or challenge assumptions.

Inside a capsule: technical background and roadmap

What saved context carries

ContextCapsules are reusable context objects, not prompts. They can hold sources, assumptions, tacit knowledge, reasoning devices, review gates, and allowed agent behavior.

Claims, sources, memories and limitations carry labels. These support review; they do not guarantee correctness or enforce every recorded preference.

AI can propose improvements and draft review cards. Human review decides which proposals enter the saved capsule. Review decisions do not independently verify a claim.

What reaches the model

An attachment records the capsule’s saved version. Sending a question resolves that version and its source references into context. Missing or archived versions require a new selection. Source document bodies and optional workspace or web research are separate dependencies, not immutable historical snapshots.

Context details offers optional User, Situation, Tool and Output roles. Ask to use only attached material when you want to limit research. Deep research adds a multi-step task with a run history to inspect.

A handover includes saved work, a bounded context projection and references. It does not transfer original-source permissions or guarantee that another tool can import it. Package hashes check integrity, not analytical validity.

Optional research

Premium web is not enabled by default. Source discovery depends on configured providers and the selected workflow. GDELT and Wikidata research signals are advisory; candidates need review before entering the saved capsule.

Related projects and roadmap

TACITUS is an independent experimental project by Giulio Catanzariti, not an incorporated legal entity. PRAXIS is its visible workbench. DIALECTICA develops the ingestion and context layer; AGON and KAIROS are upstream research references.

Shared community capsules and collaborative review are roadmap goals. Capsules should be forkable and inspectable, with provenance and review state visible. Upstream tools remain integration layers or roadmap lanes when they are not the visible production surface.

Independent experimental software. Use public or synthetic material while evaluating it. Read the privacy notice.