What is enveigh?
A native macOS secrets vault built for the age of coding agents: agents use your API keys without ever seeing them.
enveigh keeps your API keys, tokens, and connection strings in a local, encrypted vault on
your Mac, and hands them to your tools at runtime, so they never sit in a repo, a .env
file, a shell history, or an AI agent's context window.
The core idea is use, don't see: an agent (Claude Code, Cursor, Codex, …) can run your test suite, your dev server, or your deploy with every secret present as an environment variable, while the values are redacted from everything the agent reads back. The model never receives a plaintext key.
The CLI and your agents: a 37-second walkthrough
Why this exists, in 35 seconds
How the pieces fit
| Piece | What it is | Sees plaintext? |
|---|---|---|
| The app | The vault UI: secrets, environments, rotation, audit log, Touch ID lock | Yes, it is the vault |
| macOS Keychain | Where every secret value actually lives, device-only | N/A |
| The broker | A local socket the unlocked app exposes, so tools can request values | Yes (it serves them, policy-gated) |
enveigh CLI | Your terminal tool: enveigh run --env prod -- ./deploy.sh | Injects values into your command |
enveigh-mcp | The MCP server agents talk to | No, output is redacted by default |
Three properties hold everywhere:
- Values live in the Keychain, encrypted at rest, and only flow while the app is running and unlocked. Lock the app (or quit it) and every path goes dark.
- Every value access is audited: reveal, render, export, CLI, and agent access all land in a local audit log before the value is released.
- Agents get a working environment, not keys. The default agent tool runs your command with secrets injected and returns output with the values redacted.
Where to go next
- Quickstart: install, import your
.envfiles, connect your agents. - Builds & dev servers: why
npm run build/npm run devwork unchanged. - CLI reference: every command, with examples.
- MCP server: the agent tools, per-client scoping, and the reveal opt-in.
- Security model: exactly what protects what, and from whom.