Trust / Security

Know what runs.
Know what leaves.

Docker execution and external model inference are separate.

Code and provider access

The CLI can submit the selected diff, files read during investigation, generated tests, and test output to configured providers. With your own keys, inference requests go directly to those providers under your account. Their data handling and billing terms apply.

.arnxignore and secret-file/content filters reduce what is included. They are not a complete detector for confidential information. Inspect exclusions and scope. arnx allow-secret can allow particular key-shaped content, such as a fixture; do not use it for real credentials.

Setup and review exclude untracked files by default when the repository has commits. --untracked or [review] untracked = true opts them in; --no-untracked overrides the setting. With no commits, non-ignored files form the initial change. Stage and review files deliberately.

Docker boundaries

Tests execute against a copy of the repository. Test containers have CPU, memory, and process limits; capabilities are dropped and no-new-privileges is set. Runs use no network or an internal network for configured services. The test environment receives explicitly configured variables rather than the host’s entire environment.

Image builds have network access for dependency installation. Setup commands, service images, test code, and declared variables still need inspection. Docker shares the host kernel; it is not a perfect boundary for hostile code. The CLI itself also needs access to your repository, provider credentials, and Docker daemon.

Retained tests and reports

Review reports and generated tests are stored locally under .arnx/; accepted tests are copied into project test directories. .arnx/accepted-tests.json records their hashes. Review them before committing or running them outside the arnx environment.

arnx cloud

Start with $5 in Cloud credit. Tool calls, Docker tests, the before/after check, and report generation run on your machine. arnx cloud receives the diff, the files the agent reads, the tests it writes and their output, then passes them to its model providers. The service is separate from the open-source software and requires its own service and privacy terms.

Private deployment requirements

For Enterprise, contact us to discuss your required deployment location, model access, data retention, and support. Deployment and data boundaries depend on the agreed setup; using local Docker alone does not keep inference requests local. Discuss Enterprise requirements.

Report a security issue privately

Email [email protected] with “arnx security report” in the subject. See the repository security policy. Include affected versions, impact, and a minimal reproduction you can safely share. Do not put credentials or exploitable private details in a public issue. We do not currently publish a guaranteed response time or a bug-bounty program.