Checks who is asking
Every call carries a token for one named principal, and the broker reads the calling process's ancestry before anything else. An agent without a grant files a request that a human answers.
Titanwarden is a local secret broker for machines where AI agents work. An agent names a secret and a command. The broker starts the command with the value inside it and hands back only the output. The agent never holds the value.
Agents need credentials to do real work: open a pull request, deploy, back up a database. A password manager is built for a person: unlock, read, paste. Hand one to an agent and the value lands in its terminal, its transcript, its logs and the next tool call. One shared unlock then serves every agent on the machine, so one careless pane exposes everything.
Titanwarden turns that around. The command line and the agent socket have no
get. The only way to use a secret is to run something with it.
Every call carries a token for one named principal, and the broker reads the calling process's ancestry before anything else. An agent without a grant files a request that a human answers.
The audit row commits before the value leaves broker memory. If that write fails, the command does not start. Every use records who, which secret, which version and when.
The child's output passes a redactor that knows every stored value in raw, base64,
hex and percent form. A match comes back as tw:redacted.
One encrypted SQLite store per machine (AES-256-GCM values, Argon2id passphrase and recovery slots). A restart drops the keys and a human unseals.
You need macOS or Linux, Python 3.12 or newer, uv, and SSH from a device you trust. The first setup is human-only.
git clone https://github.com/frontier-infra/titanwarden && cd titanwarden uv sync --frozen && scripts/install.sh twadmin init # over SSH from your own device twadmin unseal && tw status
twadmin init asks for a passphrase twice (at least 30 characters) and prints
the recovery key once. Record it offline. Then twadmin unseal opens the vault
and the broker serves agents.
Do not let an agent run twadmin init, unseal, recovery rotation or
reveal. If you hand the install to an agent, let it run
uv sync --frozen and scripts/install.sh, then stop.
twadmin put infra/github # a human adds a value tw run --env GITHUB_TOKEN=infra/github -- gh pr create tw names infra # names and versions, never values tw who-touched infra/github # the audit trail for one secret
The full guide, agent usage and threat model are in the repository: guide · agent usage · threat model.
tw run yet.