AI-NATIVE KUBERNETES TUI

See the cluster.
Drive the response.

Operate from the keyboard, delegate bounded investigation to an agent, or connect an external assistant over MCP. Every write still stops for you.

Start flying View on GitHub

$uv tool install 'korvid[all]'

Direct Agent MCP

Real korvid, synthetic cluster. The Agent tab is a deterministic synthetic-cluster walkthrough: real read tools over a fixture, not a live model, a live cluster, or an answer-quality claim.

One cockpit. Three ways in. You stay in command.

ONE WORKSPACE

Keyboard, the embedded Agent, and external MCP clients meet in the same visible cockpit. Agent navigation moves the panes you see; optional MCP follow mirrors supported reads, while follow-off activity still surfaces as a notification.

• Resource cockpit
• Shared Agent workspace
• MCP follow

CHECKABLE EVIDENCE

Watch-backed tables, fresh describe and log reads, and bounded Agent/MCP reads land at different moments, so snapshots can differ. Successful Agent reads mint checkable citations; deterministic diagnoses expose evidence gaps.

• Agent evidence
• MCP boundary
• Diagnosis surfaces

HUMAN AUTHORITY

Best-effort, operation-specific previews may run first. Every write still waits for a fresh approval keystroke and a fail-closed audit; a failed append blocks the mutation. MCP proposals stay off by default; provider payloads are masked, while MCP disclosure remains tool-specific.

• Approval and audit
• Provider and MCP boundaries
• Architecture