FizyonAI

No. FizyonAI is a purpose-built model hosted on Fizyonops infrastructure, trained specifically for infrastructure and platform operations. Unlike general-purpose models, it receives real cluster state from Kube Hammer to produce context-aware diagnostics. Your data is never routed through OpenAI, Microsoft, or any third party.

Exclusively on Fizyonops-owned infrastructure. There are no third-party sub-processors. Data in transit is encrypted with TLS 1.3, and query data is not persisted after the session ends.

FizyonAI Cloud requires connectivity to Fizyonops infrastructure. For fully air-gapped environments, Kube Hammer supports local AI via BYOK (Bring Your Own Key) or Ollama with zero egress.

FizyonAI bills by 16,384-token context windows - not per query. A context window is a single AI conversation turn including the cluster context, your question, and the model's response. If your interaction fits within 16,384 tokens, it costs one context window. Community tier includes 5 context windows per day (~150/month) for free. Pro includes 500 per month with add-on packs available (+500 for $3/mo or +2,000 for $10/mo). Enterprise includes 2,000 per month with custom volume pricing.

FizyonAI works with any Kubernetes distribution that Kube Hammer connects to - including EKS, GKE, AKS, OpenShift, Rancher, k3s, and bare-metal clusters.

We take model accuracy seriously - especially during beta. FizyonAI is provided as-is and all AI-generated suggestions should be independently verified before execution. That said, we have built a multi-layered fairness system to protect you from paying for obvious model failures. If the AI clearly misunderstands your query, you can prefix your next message with 'nop,' to signal a misunderstanding. An automated classifier evaluates the claim - if the misunderstanding is obvious (confidence above 85%), you receive an automatic +1 context window credit back to your quota. For hallucinations - cases where the AI fabricates information about your cluster - you can prefix with 'godd,' followed by a description of what went wrong. These reports enter a human review queue. If confirmed, you receive between +3 and +10 context windows back depending on severity. We never charge you for clear model mistakes.

Every seat starts with a reputation score of 1,000. Accurate feedback reports (confirmed misunderstandings and hallucinations) increase your reputation, unlocking faster review processing and higher trust levels. There are four tiers: Contributor (1,000-1,099), Trusted Reporter (1,100-1,499), Quality Guardian (1,500-1,999), and Infrastructure Sentinel (2,000+). The system uses different verification mechanisms at different stages - low-effort signals like thumbs up/down are processed instantly, misunderstanding reports go through an automated classifier, and hallucination reports receive human evaluation. This graduated approach keeps the system fair while preventing abuse.

Kube Hammer includes a configurable pre-send redaction engine that runs entirely on your machine before any data reaches FizyonAI. It ships with 12 built-in regex patterns that detect and mask common sensitive data - including Kubernetes secret values, bearer tokens, AWS access keys, PEM private keys, JWT tokens, database connection strings, email addresses, and IP addresses. You can enable or disable any built-in pattern, and add your own custom patterns (regex or simple keyword matching) for organization-specific formats like internal tokens or project identifiers. A built-in test tool lets you paste sample text and preview exactly what gets redacted. When redaction is active, a shield indicator shows how many items were masked before each message is sent. For fully air-gapped environments, use Local AI mode where no data leaves your machine at all.

Yes, FizyonAI is currently in beta. Model behavior, response quality, and availability may change as we iterate. Beta means the model is production-capable but actively improving. We display a clear beta indicator on every AI response in Kube Hammer, and destructive commands suggested by the AI always require explicit user confirmation before execution. FizyonAI output is not a substitute for professional judgment - always review and validate AI suggestions before applying them to production clusters. We strongly recommend testing AI-suggested commands in staging environments first.

FizyonAI context windows are bundled with your Kube Hammer tier - there is no separate FizyonAI charge. Kube Hammer subscriptions follow the refund policy published at hammer.fizyonops.com/refund. Additionally, the feedback mechanisms (nop, godd) are designed to credit back context windows for verifiable model failures, so you are not penalized for the model's mistakes. If you experience persistent quality issues, contact [email protected] and we will work with you directly.