Privacy & security

Basal and the rest of Luna’s agents run on Luna’s own infrastructure, not a third-party SaaS. That means the posture is ours to set. Here’s what’s in place today.

Your conversations are private

Each Lunite gets their own dedicated agent instance with its own memory. Conversations are stored separately per user — your messages with Basal are not visible to other employees, and Basal can’t confuse your conversation with anyone else’s.

Zero Data Retention at the model layer

Basal uses Claude (Anthropic’s AI model) as the underlying language model. Luna’s Anthropic workspace has Zero Data Retention enabled. In plain terms: Anthropic does not log, store, or train on the messages you send through our API.

Luna-owned egress

All model calls flow through luna-ai-proxy, a Luna-owned service. That’s the only place API keys live. Individual agents never hold credentials, and request metadata is hashed before being sent to the AI Gateway — so the audit log shows that a request happened without exposing what was said.

Identity verification

Every message Basal receives is tied to a verified @lunadiabetes.com email address from Slack. There’s no way for someone outside the company to reach Basal through the Slack app, and there’s no anonymous access.

Audit logging

All interactions are logged through Cloudflare’s AI Gateway for security and observability. The log captures request metadata — timing, cost, which agent responded — not message contents.

What not to send

Even with these controls in place, treat Basal as an internal productivity tool, not a cleared system for regulated data. Do not paste:

• Protected health information (PHI) or patient identifiers.
• Personally identifiable information of customers or trial participants.
• Confidential third-party information under NDA.
• Proprietary competitive intelligence you’ve been asked not to share.

Questions about whether something is safe to share with an agent? Ask john@lunadiabetes.com.