Changelog

Build infrastructure agents from plain CI jobs and the Ankra CLI - no framework, no orchestration layer:

• PR review agent - validates cluster-definition changes against the live platform and comments on the PR with what will actually change, including live capacity headroom checks.
• Deploy watcher - ankra cluster apply --wait on merge, verifies rollout health, and posts a root cause to Slack when a deploy fails instead of a red X.
• Scheduled infra watcher - runs every 30 minutes, stays silent while healthy, posts to Slack only when something degrades.

ankra chat is one-shot and scriptable, with server-side access to cluster context: logs, events, manifests, and stack history. Read the guide.

Cursor is good at application code but loses context the moment a change crosses into Kubernetes, Helm, and CD pipelines. The Ankra CLI now works as an infrastructure subagent: Cursor delegates cluster-aware questions (“why is this pod crashing on staging?”) to ankra chat, which answers with grounded cluster state instead of guesses.

Includes llms.txt support on the docs site so agents can self-serve CLI usage. Read the setup guide.

AI agents at other companies have wiped real production databases this year. Ankra’s answer is architectural: the AI can only produce drafts. Every generated stack, values change, or fix lands as a draft with a full preview (what changes, what restarts, what depends on it), requires explicit human approval, and commits through GitOps with an audit trail and one-click rollback.

The AI gets the speed; Git keeps the authority. How it works.

Stacks are one object deployed to many clusters with different variables - so cloning production for local dev, QA review, or a client demo is a first-class operation, not a weekend of scripting. Works from OrbStack on a laptop to GitHub Actions to ephemeral staging clusters.

Read the 5-person-team walkthrough.