AI analysis
This page covers how AI analysis works in CLARITY and what to expect.
Built-In AI
AI analysis is included in every paid tier — Starter, Pro, Business and Enterprise all carry the same ai_insights and ai_explain entitlements. Your tier is not what determines whether AI works.
What determines it is the deployment's AI_PROVIDER setting, which your operator configures:
AI_PROVIDER | What happens | Where your data goes |
|---|---|---|
none (default) | Rule-based analysis only. No AI badges, Explain returns an error. | nowhere |
claude | Full AI analysis. Requires an Anthropic API key. | api.anthropic.com |
gemini | Full AI analysis. Requires a Google API key. | generativelanguage.googleapis.com |
ollama | Full AI analysis, self-hosted. | stays in your deployment |
vllm | Full AI analysis, self-hosted. | stays in your deployment |
AI is off unless your operator turns it on
AI_PROVIDER defaults to none. A new deployment produces no AI badges and no explanations until someone sets it and supplies a key. This is a deployment setting, not a subscription setting — upgrading your tier will not enable AI. See Data privacy before choosing a provider.
What AI adds
When AI analysis is active, it augments the deterministic rule engine with:
- Batch validation — Findings are sent to the AI for review after a manual sync or when a cloud account is first connected. Scheduled background syncs do not trigger batch validation, so badges refresh when you sync by hand, not on the automatic cadence. Results appear as confidence badges (
Agree,Modify,Disagree) on the findings the model chose to prioritise — a finding the model did not rank carries no badge. - On-demand explanations — Click Explain on any finding to get root cause analysis, impact assessment, and step-by-step remediation actions.
- Contextual awareness — AI considers your full usage history, resource relationships, and cloud best practices when evaluating findings.
Rule-Based mode
When AI_PROVIDER is none, CLARITY operates in rule-based mode:
- All insights are generated by deterministic threshold rules (e.g., CPU < 5% = idle)
- All recommendations are computed from usage metrics and pricing data
- All anomalies are detected via statistical analysis of historical baselines
Rule-based mode is fully functional — AI adds an extra validation and explanation layer on top of the same underlying analysis.
Verifying AI is working
To confirm AI analysis is active:
- Trigger a cloud data sync from the dashboard
- Wait for the sync to complete
- Navigate to any recommendations or insights page
- Look for AI validation badges (
Agree,Modify,Disagree) on findings - Click Explain on any finding to test on-demand analysis
Use a manual sync for this test — scheduled syncs do not run batch validation.
If badges do not appear, the cause is almost always that AI_PROVIDER is unset or its API key is missing. Ask whoever operates your instance to check it. Clicking Explain is the quickest test: when AI is not configured it returns "AI provider not configured" rather than failing silently. Changing subscription tier will not affect this.
Next steps
- Learn how AI analysis works in the AI Overview
- Set up cloud credentials to start syncing data