v0.1.0 — Open Source

agent-env

Universal secret injection for AI coding agents.

One command. Every agent. Secrets stay out of your shell history.

terminal
$ curl -fsSL https://raw.githubusercontent.com/jordanburke/agent-env/main/install.sh | bash
then: agent-env run claude

The problem with secrets today

AI agents need API keys. Current approaches all have trade-offs.

~/.bashrc Export in .bashrc

Leaks to every process, clutters your shell

$ KEY=... Prefix every command

API keys in shell history, tedious

.envrc Use direnv

Not designed for secrets, no encryption

N configs Configure each agent

Different config per tool, fragmented

agent-env — one secrets file, automatic discovery, optional encryption. Works with every agent.

How it works

Four steps, zero config, one bash script.

1

Discover

Walks from PWD up to git root, then global fallbacks. Finds .sops.env or .env automatically.

# auto-discovers .env or .sops.env
2

Detect

Auto-detects format: grep "^sops:" header means SOPS, otherwise dotenv. No config needed.

# grep -q "^sops:" → SOPS mode
3

Load

SOPS files decrypted via sops exec-env. Dotenv files sourced with set -a. Secrets in memory only.

# secrets loaded into env vars
4

Exec

Process replacement via exec. agent-env ceases to exist. Only your agent remains.

# exec claude "$@"

Get started

Two ways to install. Takes seconds.

One-line install

curl -fsSL https://raw.githubusercontent.com/jordanburke/agent-env/main/install.sh | bash

Clone and install

git clone https://github.com/jordanburke/agent-env.git
cd agent-env && ./install.sh

Quick start

1.
agent-env init — create .env in your project
2.
Add your API keys to .env
3.
agent-env run claude — done

Want encryption? Use agent-env init --sops for SOPS + age encrypted secrets that are safe to commit to git.