AI products with operator control, evidence, and integration at the center
Krystal Unity builds practical AI products and the shared infrastructure that lets agents operate with context, policy, memory, billing, QA, and audit trails.
The portfolio spans BlackDome, Krystal Trace, Krystal Voice, HaltState, KrystalView, FixLine, Arena, Grant Hunter, KMH minting, and a paused CashFlow Trends experiment. The important common thread is not a vague promise that AI improves everything automatically; it is a governed operating stack that can connect products, agents, customers, and evidence.
The shared KU stack behind the product portfolio
BlackDome, Krystal Trace, and KrystalView are being shaped around MCP-style interfaces so agents can query evidence, analysis, and operational state through governed access paths.
KU State, Nexus Knowledge, vector retrieval, and project catalogues give coding sessions and agent workflows a current map of products, status, patterns, and prior decisions.
Voice Hub, Billing Hub, QA Hub, admin surfaces, Redis, and the semiotic bus provide reusable services instead of each product becoming a separate one-off build.
Product work emphasizes traceable decisions, reproducible evidence packs, live endpoint checks, and explicit human review where automated systems can affect customers or operations.
Planner, builder, auditor, Sentinel, and operator workflows have evolved into a hybrid CLI and service model rather than a single monolithic app builder.
BlackDome, Arena, and related Sentinel work focus on controlled deception, training, quarantine, and evidence collection rather than unsupported claims of automatic defense perfection.
Seven related pending filings describe the governance layer behind the work.
Threshold-gated self-modification, isolated meta-builders, progressive rollout, HSM-anchored oversight, trust-domain isolation, signed capability tokens, revocation, quarantine, and tamper-evident ledgers.
Covenant-based orchestration, role-scoped tools, semantic disagreement gates, human vetoes, context-aware code generation, contract-test generation, telemetry-trained scaffolding, and rollback paths.
Memory-palace continuity, signed continuity state, proof-pack binding, and semiotic symbolic chains for typed, compressed, injection-resistant agent control-plane communication.
Products, pilots, MCP integrations, or patent-pending governance licensing discussions.