AI coding agents may process prompts, selected code, repository files, diffs, command output, account information, usage metadata, diagnostics, feedback, and data exposed through integrations. What is transmitted, stored, or shared depends on the product, model route, feature configuration, and organization settings.
Audit the data path by category:
- Task content: Prompts, attachments, code context, test output, and generated changes.
- Account data: Identity, organization membership, device, and authentication records.
- Operational data: Feature use, errors, tool calls, performance, and security events.
- Integrations: Repository, issue, chat, database, or deployment data available to connected tools.
- Feedback: Conversations or examples intentionally submitted for support or improvement.
Ask separate questions about processing, transmission, indexing, retention, deletion, and model training. Those terms are not interchangeable. Also identify subprocessors, regions, administrator controls, and optional telemetry. BYOK may change the model-provider route but does not automatically remove the agent platform from the path.
For Verdent specifically, the current Privacy Policy describes input/output, account, usage, and optional indexing data, while the Security Policy identifies infrastructure and inference subprocessors. Other products require their own current documents. Evaluate with synthetic data first and keep secrets outside agent-readable paths.
Related reading: Is my code safe with AI coding tools? and Does BYOK improve coding-tool privacy?.
