Codex sends its requests to BlindHush as a custom model provider. The proxy masks your instructions, messages, shell commands and output, patches and tool calls, then forwards them to OpenAI with your own OpenAI API key.
Set up
The installer adds this to ~/.codex/config.toml (and backs up the old file):
model_provider = "blindhush"
[model_providers.blindhush]
name = "BlindHush (masks personal data before OpenAI)"
base_url = "http://127.0.0.1:8788/v1"
env_key = "OPENAI_API_KEY"
wire_api = "responses"
Codex uses OPENAI_API_KEY from your environment.
Limits
- This works when Codex uses an OpenAI API key. Signing in with a ChatGPT account sends requests to a different endpoint, which BlindHush does not cover.
- Encrypted reasoning items pass through untouched (they are not readable text).