From Prototype to Local AI Desktop Platform

RELDEN is not another AI model. It is the environment that decides how AI should be used. The desktop coordinates requests through a provider-neutral core, so the product is not tied to one model or vendor.

Complete and verified

The first four development phases established the current desktop foundation:

• A typed execution layer carries each request through the application with a request ID, cancellation and safe failure handling.
• Conversations and messages persist in local SQLite storage and remain available after the application restarts.
• Supported local models run through llama.cpp and stream responses incrementally into the desktop interface.
• A deterministic router evaluates privacy, model availability, required capability, context capacity and user preference before selecting a permitted execution path.

This means RELDEN Desktop now has a functioning local-first request flow: a request enters the execution layer, is evaluated by the router, reaches the local provider and streams back into a persistent conversation.

Local-first by operation

Current production execution is local-first. Supported AI workloads run on the user’s machine, and conversation history is stored in the local desktop database.

The router can enforce a local-only policy. If no permitted execution path exists, the application can block the route or require clarification instead of silently sending work elsewhere.

The provider architecture already separates RELDEN Core from a specific AI vendor. That separation is important because it allows local and future cloud providers to expose typed capabilities without becoming hard-coded product dependencies.

In development

Production cloud execution is intentionally disabled. The cloud provider architecture exists, but secure OS-native credential storage must be completed before cloud requests are permitted in production.

The current development phase also includes controlled tools, Memory refinement, the Learning Log, settings refinement, security hardening, UX refinement and acceptance testing.

The Memory runtime and contextual conversation foundation already exist. The complete persistent-memory experience, review controls and richer provenance are still being developed. RELDEN does not yet provide autonomous agents as a product capability.

Planned

Voice remains planned. It will follow the same product principles: explicit capability boundaries, understandable state and user control.

RELDEN-8B continues as a separate model research workstream. RELDEN Desktop does not depend on RELDEN-8B; the provider-neutral architecture allows supported local models to evolve independently from the desktop orchestration layer.

What comes next

The focus is now on completing the orchestration layer around the working local foundation. Secure optional cloud support must respect the same privacy and routing policies as local execution. Controlled tools need clear permission boundaries. Memory needs reviewable user controls. The Learning Log needs to make important execution outcomes understandable.

These are product and safety foundations rather than headline features. Completing them is what turns the verified local desktop path into a dependable wider AI environment without overstating what the current private alpha can do.

LST — Laubenstein Technologies