Skip to content

FAQ

No. kovra is a local tool — the vault lives on your machine and nothing is transmitted as a side effect of normal use. The only network calls happen when you use a cloud reference (kovra resolves it under your own provider identity) or deliberately share a package. There is no telemetry and no phone-home.

Yes. Everything except cloud references (which, by definition, call your cloud provider) works with no network at all.

By default, nothing. A secret only moves when you explicitly share it (a sealed package, encrypted to the recipient) or when a cloud reference resolves against your provider. Even then, the plaintext is never written to disk, argv, or an agent’s context.

Yes — that’s the point. It holds addresses, not values. A leaked .env.refs exposes where secrets live, never the secrets. Add a git hook as a backstop against committing real values by accident.

It can use them, not see the sensitive ones. An agent over MCP runs under a scope and never receives the plaintext of a high, prod, or inject-only secret. The only thing it can read back is an ordinary secret you explicitly marked revealable.

In an encrypted vault under ~/.vaults (or KOVRA_VAULT_DIR). Every entry is sealed; see Configuration and Cryptography.

It’s source-available under the Business Source License 1.1, and each version becomes Apache-2.0 four years after release. See License.

No. It’s a local CLI. The Web UI is on-demand and loopback only — it isn’t exposed to the network and shuts down when idle.

Only to use kovra from an AI agent. The CLI and vault work on their own; kovra-mcp is the optional bridge for Claude Code and other MCP clients.

Restore from a key backup — see Backup & recovery. Make that backup before you need it.

macOS on Apple Silicon is the reference platform today. Windows (Windows Hello + Credential Manager) and Linux are on the roadmap.