nano SIEM
Settings

AI & pivt

Supported in:Self-hosted

Turn pivt on or off, choose the AI tier and tuning mode, give pivt organizational context, and track AI usage.

Use Settings → AI & pivt (/settings/ai) to configure nano's AI assistant and review usage. pivt powers the parser, query, detection, tuning, dashboard, and search-summary wizards, notebook chat, and automated investigations.

pivt is included with nano Self-run. The open-core build doesn't include it.

The page has six tabs:

TabWhat it holdsPermission
ProvidersProvider API keys and enablementsettings:ai_providers
ModelsModel catalog and AI tuning modesettings:agent_models
Agent ModelsPer-agent model overridessettings:agent_models
GuidanceOrganizational contextsettings:ai
UsageAI usage and creditssettings:ai
MonitoringProvider health checkssettings:ai

Set up a provider

You bring your own AI provider:

  1. Providers: add credentials for a provider (Anthropic, Google AI, OpenAI, Azure OpenAI, AWS Bedrock or Cloudflare Workers AI), enable it, and click Validate. Keys are stored encrypted.
  2. Models: sync and review the model catalog from your enabled providers.
  3. Agent Models: optionally assign a different model to individual agents.
  4. Guidance: add priority threats and custom instructions.
  5. Monitoring: turn on provider health checks to be notified when a provider stops responding.

AI tuning mode (Models tab):

  • On (default): automatic tuning proposals plus the Tune wizard.
  • Manual: the Tune wizard only; no automatic proposals or notifications.
  • Off: no AI tuning; the Tune wizard is refused.

The mode applies together with each rule's own auto-tuning switch.

Model calls go through a Cloudflare AI Gateway that you set with CLOUDFLARE_AI_GATEWAY_URL (and CF_AIG_AUTH_TOKEN if the gateway requires authentication). If the gateway URL isn't set, AI stays off unless an enabled provider points at an on-premises OpenAI-compatible endpoint through its base URL. That is how air-gapped installs run AI.

Product-docs grounding

When you ask pivt a product question ("how do I…", "what is…"), it can ground the answer in the public nano documentation and cite the page it used. It retrieves from nano's docs search service (https://docs-search.nano.rs), which indexes public documentation only and never your data. The text of your question (up to 1,000 characters) is sent as the search query. Retrieved passages are treated as reference material, never as instructions.

  • Self-run and open-core installs: off. pivt answers product questions from its model alone. The docs search token is a nano server-side credential and is not issued to self-hosted installs.
  • Air-gapped installs (AIRGAP_MODE=true): always off, with no outbound call.

To force it off on a deployment that has a token, set DOCS_SEARCH_URL=off.

Permissions

PermissionAllows
settings:aiGuidance, Usage and Monitoring
settings:ai_providersManage AI providers
settings:agent_modelsManage the model catalog, agent models and tuning mode
melod:chat, melod:query, melod:parser, melod:detection, melod:summarize, melod:dashboard, melod:notebookUse the matching pivt features

Troubleshoot

Need help with these checks? Contact support with the request ID and time with timezone.

AI stays off

Check CLOUDFLARE_AI_GATEWAY_URL and the enabled provider. Without a gateway, an enabled provider must point to an on-premises OpenAI-compatible endpoint.

What's next

On this page