BlogReleasesIssuesDocsDownloadsStatusLooking GlassContact Sales

Codex

OpenAI Codex runs shell commands, so it uses Clavitor the same way any terminal agent does: through clavitor-cli.

1. Enroll the CLI

Follow agent setup: create a CLI client in the vault, named for example "Codex", scoped to the entries Codex may use. Then, on the machine where Codex runs:

clavitor-cli init

Paste the blob, then the key.

2. Give Codex the instructions

clavitor-cli skill install knows only Claude Code and OpenClaw, so for Codex you add the instructions yourself. The CLI prints them:

clavitor-cli skill print

Codex reads project instructions from AGENTS.md. Append the printed text there:

clavitor-cli skill print >> AGENTS.md

The instructions tell the agent to fetch one credential at a time, only when it needs it, to find the exact title with clavitor-cli find, and never to make a secret the output of a command. Claude Code summarises them.

What it looks like

clavitor-cli find openrouter
curl -H "Authorization: Bearer $(clavitor-cli get OpenRouter --field key)" https://openrouter.ai/api/v1/models

For a config file with several secrets, Codex uses render and pipes the result into the program:

clavitor-cli render app.config.json | myapp --config -

Limits worth knowing

  • Codex can read only the entries in the client's scopes, and only L1 and L2 fields. L3 fields stay closed.
  • The read quota applies, and a client that goes over its hourly limit twice within two hours is locked until you unlock it.
  • The instructions are guidance for the model. What the vault enforces is the scope, the quota, the IP pin and the refusal to release L3.