Lorenz combines an executive workspace, source-grounded Second Brain, durable agent control, jurisdiction-aware routing, and evidence-backed actions—with preview and activation gates shown explicitly.
Interface, knowledge, execution, and model infrastructure remain connected by tenant, policy, evidence, and recovery.
Every workload can be assigned to a local-only, in-country, or approved external lane based on data class and regulatory scope.
Retrieval combines semantic, lexical, and operational memory layers while the RAG Governor handles claims, conflicts, and memory coherence.
Prompts, commands, and risky parameters are screened for destructive intent, prompt injection markers, and dangerous execution patterns.
Skills and tools run under explicit policy, human confirmation, workspace boundaries, and action logging instead of broad ambient trust.
Action Journal, undo snapshots, heartbeat, and version traceability make incidents visible and reversible.
Capability maturity
Lorenz publishes status at capability level. Source code, controlled preview, tenant activation, and production proof are not interchangeable.
Bind work to a user, tenant, role, policy, approval, audit record, and recovery path.
Current evidence
Tenant context, forced RLS in critical domains, Guardian, Action Journal, DLP/risk gateway, and work packets are implemented.
Market context
Premium suites are strong on identity and administration; open-source products range from personal defaults to enterprise ACL editions.
Compile governed sources into citable pages with revision, conflict, and explicit degradation state.
Current evidence
Second Brain is implemented; contextual Retrieval v2 has signed qualification but remains tenant-activation gated.
Market context
RAG is common across the market; source compilation, conflict lineage, and fail-closed activation are Lorenz's sharper claim.
When the controlled preview is activated, choose Auto, Quick, Think, Research, or Create through server-owned policy while unqualified modes remain visibly blocked.
Current evidence
Backend, web, and iOS contracts plus Routing Receipt and Context Manifest are implemented; Analyze and Build remain unavailable and DEV activation is open.
Market context
Many assistants expose model or mode selectors. Lorenz focuses on making the resolved route, source use, egress, cost band, and degradation inspectable.
Bind sources, claims, citations, runtime identity, results, and completeness to content-addressed evidence.
Current evidence
Research Evidence is an accepted backend foundation; the fixed Analysis package and graph exist locally but live signed sandbox qualification is pending.
Market context
The differentiation is not a longer answer: it is a reproducible boundary between complete, partial, insufficient, qualified, and unqualified results.
Supervise long-running agent work through immutable revisions, leases, checkpoints, lineage, budgets, and controls.
Current evidence
The graph kernel, mTLS runner protocol, web/iOS supervision, SDKs, and Project Agent vertical slice are source-verified alpha.
Market context
Builder platforms and premium clouds are rapidly expanding orchestration; Lorenz focuses on evidence and operator authority.
Move from planning and cited research to fenced changes, tests, review, approval, and draft-only publication.
Current evidence
Repository reconnaissance, test evidence, writer locks, approval binding, signed commits, and idempotent draft PR effects are implemented in the alpha graph.
Market context
Coding agents often optimize for delivery speed; Lorenz optimizes for a reviewable authority and evidence chain.
Review same-repository pull requests on the independent runner without granting approval, merge, or branch-write authority.
Current evidence
Signed webhook intake, exact base/head binding, secret scanning, changed-line validation, deduplication, and a publication kill switch are implemented; the real DEV canary remains open.
Market context
Coding platforms lead on delivery integrations; Lorenz's narrower claim is review evidence tied to tenant policy, runner identity, and revocable publication authority.
Resolve immutable skill versions with declared data, egress, secret, effect, approval, and sandbox boundaries.
Current evidence
Manifest lifecycle and prompt/document packages are implemented; executable packages require separate qualification.
Market context
Open ecosystems lead in extension breadth. Lorenz deliberately trades installation freedom for provenance and policy.
Keep voice continuous across web/mobile while selecting an approved premium or private runtime under consent.
Current evidence
Streaming voice and account-scoped profiles are available; the packaged MINT multilingual clone still needs exact-hardware qualification.
Market context
Premium assistants set a high UX bar; OpenClaw sets a high channel-breadth bar. Lorenz adds explicit routing and consent boundaries.
Use corporate identity and permission-aware sources without turning email domains into tenant authority.
Current evidence
Microsoft tenant binding, admin consent, strict token verification, and separate mailbox/SharePoint permissions are implemented but activation-gated.
Market context
Microsoft, Google, Glean, and Salesforce lead through native ecosystem depth; Lorenz competes on neutrality and explicit boundaries.
Observe approved network appliances through tenant routes and content-minimal AI results without granting mutation authority.
Current evidence
The read-only broker, route registry, Passbolt references, quotas, policy hashes, and System Doctor states are implemented; live canary remains open.
Market context
This is an operational control-plane extension rather than a generic connector count claim.
Three pillars that define how Lorenz behaves as a governed platform instead of a generic agent shell.
Lorenz can run where the customer needs the data to stay: on-premises, in-country, in-region, or inside a private perimeter aligned with local law.
Security is not a plugin. Sensitive routing, AEGIS controls, scoped execution, audit trails, and rollback are built into the operating model.
Lorenz is designed for governed execution, advanced RAG, and multi-tenant control instead of ad hoc tool sprawl and silent data egress.
Lorenz does not treat retrieval as a black box: memory, OCR, and ranking stay under policy and observability.
Isolation does not live in one control only: it is enforced across identity, queries, workspaces, runtime, and recovery.
Lorenz is built to act for real, but never under ambient trust: skills, tools, and side effects stay inside explicit boundaries.
Start deploying AI operations with built-in sovereignty, security, and auditability.