On this page
Dranz is a coding-agent workbench, not a safety guarantee. It helps you inspect routes and approve consequential actions. Approvals are a review step, not a sandbox. You are still responsible for the project, the provider you select and the commands you allow to run.
Keep a human in the loop
Treat model output as a proposal, not a source of truth. Read diffs, review commands, run tests, check dependency licenses, and verify the result against what you asked for. Do not let Dranz be the only reviewer for code that handles money, health, identity, safety, security, access control or other high-impact decisions.
The default permission mode is Accept edits: shell commands and other consequential actions pause for approval. If you choose a less restrictive mode, do it deliberately. A model can make a mistake at machine speed.
Protect secrets and sensitive data
- Keep API keys, passwords, private keys, tokens and recovery codes out of prompts, issues, screenshots, support messages and committed files.
- Do not send personal, confidential, regulated or third-party data to a provider unless you are authorized and the selected route is appropriate.
- When inference must stay on your machine, use a local model route (a Dranz Pro feature), and still inspect any network-capable tool you approve.
- Keep independent backups before allowing an agent to edit or delete files.
The provider badge tells you where inference went. It does not mean every network operation was local: web access, MCP servers, registries, update checks and approved shell commands can create separate routes.
Verify providers and downloads
Venice, direct providers, local model services and tools have their own terms, retention policies, pricing and security properties. Read them before connecting an account. Dranz does not guarantee a third party's availability, output, retention or security.
Download Dranz only from dranz.ai, verify the architecture, and follow macOS's security prompts carefully. Early builds can contain defects, break compatibility, lose data or expose security weaknesses. Do not install a build on a machine or in a project where that risk is unacceptable.
Report a security issue
Do not publish credentials, exploit details or private user data in a public issue. Send a short report to legal@dranz.ai with the affected version, reproduction steps and a safe contact method. We cannot promise a response time or a particular remedy, but a private report gives us a chance to protect users before details spread.
For the contract terms and full risk allocation, read the Terms of Use.