faq.md

FAQ

Does it need an API key?

No. The deterministic engine works fully offline with the packaged knowledge base. API keys are only for optional LLM enrichment. See AI providers.

Will it run commands or change my infrastructure?

Never. The toolkit is read-only: it reads text and produces guidance. Diagnostic commands are suggestions you choose to run; fixes are described, never auto-applied. See Security.

Is it deterministic?

Yes — the core engine produces the same output for the same input. Optional LLM enrichment is the only non-deterministic part, and it's off by default. See Comparison.

What can it analyze?

Logs, YAML/Kubernetes manifests, Terraform, Docker, OpenStack, Linux/systemd, databases, message queues, and more — plus raw command output and error strings. See Supported technologies.

How is this different from asking ChatGPT/Claude?

It's deterministic, offline-first, read-only, and its findings are sourced from auditable signatures. It can also use an LLM — but only as opt-in enrichment on top of the deterministic result. See Comparison.

CLI, SDK, or API — which should I use?

They all run the same engine. Use the CLI for ad-hoc triage, the SDK to embed analysis in your tools, and the REST API to share it as a service. See Architecture.

Which Python version?

Python 3.12+. See Installation.

How do I add support for an error it doesn't know?

Add a YAML signature — no engine code required. See Knowledge base.

Can I use my own LLM provider or a local model?

Yes. Register a custom provider with register_provider(), or point the built-in Ollama adapter at a local model for fully private enrichment. See Plugin guide and AI providers.

Does enrichment send my logs to a vendor?

Only if you enable it. When enrich=True and a provider is configured, a bounded slice of input is sent to that provider. Keep it off, or use local Ollama, to avoid any egress. See Security.

What happens if an LLM call fails?

The engine logs a warning and returns the deterministic result. Analysis is never broken by a provider error.

Is the output machine-readable?

Yes. Add --json on the CLI, or use the SDK/API to get the AnalysisResult model directly. See Output format.

Is there a hosted version?

Yes — the AI incident-response assistant offers a managed experience.

How do I keep up with new signatures and features?

Subscribe at https://devopsaitoolkit.com/newsletter and watch the repository.

What's the license?

MIT.

What's on the roadmap?

A Web UI, VS Code extension, GitHub Action, MCP server, and Desktop app — all built on the same engine. See the Roadmap.