Ask KubeOn
Ask why spend changed. Get an answer you can check.
Ask KubeOn answers questions about spend, idle, efficiency, budgets, anomalies, savings and agent health. It reads the same data you see, with your permissions, and checks every amount and percentage it writes against that data.
How it answers
Ten read-only tools, called with your permissions
To answer, the assistant calls the same APIs the product uses: cost overview, Kubernetes costs, cost changes, cloud costs, efficiency, savings, anomalies, budgets, cluster health and change proposals. It runs at most 8 rounds per question and links to the page behind every result.
- The tools only read; it cannot change anything
- Team scoping applies to the assistant like everywhere else
- Topics limited to cost, usage, budgets, anomalies, savings and agent health
Figure tracing
Every number in the answer is traced to the data
After each answer, KubeOn matches every amount and percentage against what the tools returned, allowing for rounding and simple sums and differences. The answer shows how many figures were traced, and names any that were not.
- A count of the figures traced to KubeOn data
- Untraced figures called out by value
- Thumbs up or down feedback for admins to review
In context
Explain buttons where the questions come up
Budgets, anomalies and change proposals carry an Explain with AI button that opens the assistant with the right question already asked: where a budget's spend is going, what drove a spike, whether a change is safe and what to watch after it ships.
- Context passed with the question
- Conversation history, readable only by its owner
- A weekly digest posted to your channels on Mondays
Where the Kubernetes savings are
off the monthly bill: $21,842 a month found at our sample company
Sample company, not a customer result
Model and limits
Runs on the model endpoint you choose
An admin chooses Amazon Bedrock, Azure OpenAI or Google Vertex AI as the model endpoint. The assistant is off until an admin turns it on, and each call is recorded in a usage ledger with tokens, cost and latency.
# Settings > Assistantai: enabled: true provider: bedrock # bedrock | azure-openai | vertex-ai model: anthropic.claude-sonnet maxToolRounds: 8 maxOutputTokens: 2048 questionsPerUserPerDay: 50 weeklyDigest: day: monday hourUtc: 13 channels: [slack-finops]An assistant built for numbers people check
Read-only
It can explain and point to changes, never make them.
Traced figures
Every $ and % matched against tool output.
Your permissions
Answers only cover what the asking user can see.
Suggested questions
Start from five common questions or ask your own.
Weekly digest
A summary of spend, anomalies and savings every Monday.
Feedback and usage
Admins see feedback, questions per user and model cost.
Ask KubeOn: your questions answered
KubeOn
Keep exploring
Evaluate Ask KubeOn with your team
Bring your platform and FinOps priorities. We will walk through the data, controls and evidence for your evaluation.