AI administration overview
AI Control Center
Section titled “AI Control Center”The AI Control Center governs the AI capabilities built into RezolveApp — Employee Center AI Assist and its skills. It’s organized into an overview plus dedicated tabs for skills, model providers, guardrails, and budget.
Overview
Section titled “Overview”The overview tab summarizes the current state at a glance: plugin status, a skills-status breakdown (how many skills are active, out-of-box, or not started), a “Needs attention” count, a journey checklist for setup, and a skill-usage chart of actions over time. It also surfaces the current AI routing status — whether native providers or an external gateway currently own outbound AI calls (see below).
Skills
Section titled “Skills”The Skills tab lists every AI skill (for example incident summarization, resolution-notes generation, work-notes generation, case summarization, knowledge-article generation, and more) grouped by category, with a status chip per skill — Not started, Out-of-box, or Custom — and an Activate/Deactivate toggle.
Model providers
Section titled “Model providers”The Model providers tab is where you choose which LLM provider and model serve AI requests:
- Choose a primary provider and model, and optionally a fallback provider/model.
- Available providers include the built-in RezolveApp LLM, OpenAI, Azure OpenAI, and Anthropic Claude (via AWS Bedrock) — each with its own set of selectable models.
- For providers that need an API key, enter it here: it is encrypted in the database, never displayed back in plain text. Non-secret settings (model ids, endpoints, organization ids) are edited alongside it. A Test connection action confirms the credential actually works.
- A live probe shows which provider is actually being used for AI calls right now.
Guardrails
Section titled “Guardrails”The Guardrails tab edits the same enforced guardrail policy used at runtime and shown on the AI Assist Governance page — there’s exactly one guardrail editor, not a duplicate. It covers categories such as toxicity, prompt-injection, and data-leakage detection, each with its own enabled/disabled state and action (for example block or flag). A separate, clearly-labeled “advisory settings” section below it holds additional display-only settings that are not enforced by the guardrail engine.
Budget
Section titled “Budget”The Budget tab shows month-to-date AI spend against a configured monthly limit, lets you set one or more alert thresholds (percentages of the limit), and toggle whether crossing a threshold notifies admins. Setting the limit to 0 tracks spend without enforcing a hard cap; when a hard limit is set and exceeded, live generation pauses until the next month or until the limit is raised.
External AI gateway
Section titled “External AI gateway”Instead of routing AI calls through RezolveApp’s native providers, you can point all outbound AI traffic at your own AI gateway (LiteLLM, or any OpenAI-/Anthropic-compatible proxy). Configure it from the Model providers area:
- Turn on gateway mode and set the gateway’s base URL.
- Enter the gateway’s API key — it goes into the same encrypted vault as the native provider keys, never anywhere else.
- Optionally map RezolveApp’s model ids to your gateway’s own virtual model names; unmapped ids pass through unchanged.
- Use Test connection to confirm the gateway responds.
Once gateway mode is active, your gateway becomes the system of record for provider routing: the native Guardrails and Budget tabs switch to read-only, since the gateway now owns guardrails and spend limits — any governance you want (guardrail policy, spend limits, usage logging) is configured on the gateway itself, not in RezolveApp. A half-configured toggle (enabled but no base URL) fails safe back to native routing rather than silently going nowhere.