Some AI coding tools transmit, index, cache, log, or retain source-code content; others keep more processing local. The correct answer requires the current product, feature, model route, account terms, and administrator settings.
Separate six questions:
- Does code leave the device when context is assembled?
- Is a repository index created locally, remotely, or both?
- How long do prompts, files, and tool output remain available for a task?
- Can snippets, filenames, commands, or errors enter logs and telemetry?
- What is retained after the task, and how is deletion handled?
- Can input or output be used for model training, and is consent or opt-out available?
For Verdent, the current Privacy Policy says input/output may be stored in user history, model training requires explicit consent, and optional code indexing stores embeddings rather than original plaintext code. Its Security Policy describes requests passing through Verdent infrastructure to named inference providers. These are product-specific facts, not a rule for every tool.
Review the Verdent Privacy Policy and Security Policy before using proprietary code. For another tool, inspect its own privacy, security, subprocessor, retention, and enterprise-control documents. Test with non-confidential code first, grant minimum repository access, and keep secrets outside agent-readable paths.
Related reading: Is my code safe with AI coding tools? and Does BYOK improve coding-tool privacy?.
