Local Codex for Enterprise is an experimental community fork and is not affiliated with, endorsed by, or supported by OpenAI.

No public stable release is supported yet. Security review is focused on the upcoming v0.1.0 release candidate.

Until this project has a public vulnerability disclosure program, please report issues through the GitHub repository owner using a private channel. Do not open public issues containing exploit details, secrets, private paths, tokens, screenshots with private data, or working proof-of-exploit material.

If the issue is in upstream OpenAI Codex rather than this fork, follow OpenAI's upstream security process for the original project.

The enterprise server is responsible for:

• Authentication.
• Seeded RBAC authorization.
• Workspace allowlisting.
• Worker supervision.
• Handoff token issue and consume.
• Trace-aware audit events.
• Execution receipts.

The enterprise server must not persist:

• Private/runtime prompts.
• Model outputs.
• Auth headers.
• Handoff JWTs.
• Plaintext passwords.
• API tokens.
• Repository credentials.
• Environment secrets.
• Private real-life examples.

Receipts are evidence, not reasoning. They record IDs, hashes, outcomes, and redacted metadata only.

Synthetic reusable prompt templates may appear in Context Pack examples. They must not be confused with runtime prompts, private prompts, transcripts, or model outputs, and they must not be stored in receipts or audit metadata.

This project is designed for self-hosted local or private-network evaluation. Do not expose the enterprise server directly to the public internet without external TLS termination, network access controls, secret rotation, and an independent security review.

Run at least:

just test -p codex-enterprise-server
just fmt-check
git diff --check
gitleaks dir . --verbose --no-banner
git grep -nE '<private-path-or-secret-patterns>'

Review all matches before publishing.