Your Second Brain, Your Model Provider
Choose hosted, local, or compatible model providers at runtime while keeping Arivu useful without AI.
The model connected to your second brain should be your choice. Arivu now supports Gemini, Anthropic, OpenAI-compatible services, and local or custom-compatible endpoints through runtime Admin Settings.
You can change the provider without rebuilding the app or editing deployment files. Admins choose a preset or compatible endpoint, select the model, and add credentials when that provider requires them. Local services can be keyless where supported.
One Workflow, Different AI Layers
Hosted providers can offer stronger generated summaries and synthesis. A local endpoint can keep more processing inside infrastructure you control. An OpenAI-compatible endpoint can connect another service that follows the expected interface. The right choice depends on your privacy, quality, cost, and operational needs.
Provider choice does not make every AI capability interchangeable. Models differ, compatible services may support different behavior, and embeddings or image OCR may still rely on Gemini where those features apply. Arivu makes the general text provider selectable without pretending all providers are identical.
Credentials are controlled from admin-only settings and stored for the instance. Changing a runtime provider takes effect without restarting the running service. Changes to environment-only infrastructure settings, source code, or a locally rebuilt binary still require the normal restart or deployment process.
What this means for you: You can test a hosted model, move general text generation to a local service, or disable provider use without redesigning your knowledge workflow.
Local Is the Baseline, Not a Failure Mode
Arivu remains useful with no model provider configured. Capture, annotations, notes, research objects, Inbox, Focus, Review, search, tasks, reminders, imports, exports, and deterministic enrichment continue to work.
Deterministic summaries are not presented as model-generated analysis. They provide predictable local structure when provider output is unavailable. If you enable a provider, generated text enriches that workflow rather than deciding whether the product functions.
That distinction matters for a self-hosted second brain. Your archive should not become inaccessible because an API key expires, a provider is unavailable, or you decide not to send content to an external model.
It also gives teams a clearer migration path. Start with deterministic processing, add a provider when a workflow benefits from generated text, and change that choice later without moving the underlying notes or sources. Your capture and review habits do not have to follow a model contract.
Read AI Assistant documentation for behavior and limits, or see Environment Variables for deployment defaults.
Your knowledge stays in one workflow. The optional model layer can change when your needs do.