When a pod keeps failing, the answer is usually spread across its status, its logs, the cluster events and the deployment config. This agent collects all of it in one click and hands it to an AI to reason over.
Example
On a test cluster with three broken apps, it found each problem separately: a missing DATABASE_URL crashing the payment service, an image tag that doesn't exist, and an analytics worker in a crash loop with empty logs. It gave kubectl commands for the first two, and said its confidence was lower for the third because the evidence was thin.
How it's built
Five inspectors talk to the Kubernetes API and gather their findings into one evidence report. A prompt builder passes the report to an LLM through OpenRouter (Claude, GPT or DeepSeek), and the agent returns the root cause, a fix such as a kubectl command or a YAML change, and a confidence score. InsForge handles auth, investigation history and the live progress updates in the dashboard.
