Private AI

Ask over confidential data β with the model your security team will approve.
Public chatbots send prompts to someone elseβs cloud. Private AI on Diode is the opposite: you plug self-hosted or partner-hosted models into Diode Nexus (product brand: Protected AI), then wrap them in an organizationally aware package β tools, channels, ranking, and Diodeβs end-to-end encrypted path.
The pitch is not βpick another SaaS LLM.β It is: bring the inference your policy allows, then ground it in your systems.

What this means
- Your models first β Ollama, vLLM, private GPU stacks, or a partner-hosted OpenAI-compatible endpoint. Rank them above public providers; keep cloud LLMs as optional fallbacks.
- Organization-aware tooling β MCP and tools so answers come from your registries, APIs, and files β not training cut-off alone.
- Protected agentic chat β Natural-language conversations with pipeline visibility while tools run.
- Traffic on Diode β Prefer Perimeter Discover or Diode binds so model and tool hops stay on your ZTNA / private path β not a naked public API call.
- Meet people where they work β Same package answers Diode Collab channels, web embeds, ACP for apps (including Vibe), and an OpenAI-compatible API (alpha).
Advantages
- No requirement to put confidential prompts on a multi-tenant public chatbot
- Security and compliance choose the model; operators still get a real chat + tool product
- One Nexus tenant serves Collab, Vibe coding, and custom apps β instead of every team wiring its own LLM
- Pipeline and tool activity is visible to operators, not a black box
- Works with existing Diode identities, zones, and perimeters
Applications
- Internal Q&A over policies, procedures, and operational data
- Supplier and extended-team questions answered inside a Collab zone
- Organization-aware coding and generation from Diode Vibe
- Plant / region inference that must stay on approved GPU or partner-hosted Private AI
- Embeddable help on private or protected Diode sites
- Agent APIs (ACP) so other apps call your Protected AI instead of inventing their own