Skip to content

What is Kristal?

Kristal is a framework for producing and operating deterministic knowledge artifacts (“Kristals”) that can be validated, distributed, and queried reliably—even when offline.

A Kristal is not “a database”. It is a portable artifact with: - a stable identity, - explicit inputs and build metadata, - integrity and (optionally) trust signals, - and predictable query behavior once activated.

Why Kristal exists

Most knowledge systems fail in at least one of these ways: - results depend on hidden runtime state (non-deterministic) - you cannot reproduce the same artifact from the same inputs - you cannot verify what you received (integrity/trust gaps) - you cannot safely roll forward/backward across versions

Kristal is designed to make these properties first-class: - deterministic builds (reproducibility) - content-addressed identity - fail-closed verification when declared - operationally safe distribution (activation/rollback)

What a Kristal enables (high level)

  • Build a Kristal from upstream sources
  • Validate it with schemas/profiles
  • Publish it as a versioned artifact
  • Activate it into a runtime environment (locally or in services)
  • Query it with declared capabilities
  • Evolve it safely (patches, rollbacks, compatibility rules)

Core building blocks

Artifacts

Kristal workflows revolve around a small set of artifacts you can store, move, and verify: - Exchange - Runtime Pack - Validation Report - (Optional) Shard Manifest, Federation Manifest, Authority Registry

See: Artifacts

Workflows

Workflows describe how teams operationalize Kristal: - Build & Validate - Publish & Distribute - Activate, Rollback & Downgrade - Subsets (Recipes) - Federation & Curation

See: Workflows

Concepts

These concepts show up everywhere: - Identity & determinism: what “same inputs → same artifact” means in practice
See: Identity & Determinism - Trust & authority: who can sign, how trust roots are used, and what “fail-closed” means
See: Trust, Authority & Signatures

Where to go next

Technical details

This wiki intentionally avoids repeating full specifications. For exact schemas and normative rules, refer to the technical docs in the main repository.