Status: planned / design note. A future surface, not yet released. It would be a thin client over the REST API or a bundled CLI — no new analysis logic.
Bring read-only analysis into the editor, where operators already read logs and edit manifests. Right-click a log, a selection, or a manifest and get ranked root causes, read-only diagnostic commands, and suggested fixes in a side panel.
- Analyze file / selection — command-palette and context-menu actions that send the active file or selection to the engine.
- Inline manifest validation — run
validateon YAML/Kubernetes/Terraform files and show issues as editor diagnostics (squiggles), usingValidationIssue.line/path. - Explain on hover/lookup — select an error name (e.g.
CrashLoopBackOff) andexplainit. - Results panel — a webview rendering the
AnalysisResultsections; diagnostic commands are copy-only (never executed) to preserve the read-only guarantee. See Security. - Terminal output capture — analyze the last command's output.
| Setting | Description |
|---|---|
devopsAi.mode |
cli (bundled) or api (remote endpoint) |
devopsAi.apiUrl |
REST API base URL when in api mode |
devopsAi.enrich |
Enable LLM enrichment |
devopsAi.provider |
Provider name |
Provider keys come from the environment / OS keychain, never stored in settings. See Configuration.
flowchart LR
VSC["VS Code extension"] -->|file / selection| Client["client (api or cli)"]
Client --> Engine["AnalysisEngine<br/>(REST API or bundled CLI)"]
Engine --> Panel["Results webview<br/>(AnalysisResult)"]
Like every surface, it adapts I/O only — the shared engine does the work, so results match the CLI, SDK, and API.
Suggested diagnostic commands are shown as copyable text and never run by the extension. Fixes are presented as guidance/snippets the developer applies deliberately.
Interested? Open an issue (see Contributing) and follow https://devopsaitoolkit.com/newsletter.