Confidentiality is the product, not a setting.
C2Stream is built for the three questions every attorney asks before letting AI near client email: is it confidential, could it commit malpractice, and will it sound like me. Here’s how the first two are answered in the architecture itself.
Redacted before any AI call
Personal identifiers — SSNs, payment-card and account/matter numbers, mailing addresses, dates of birth, emails, and phone numbers — are stripped from the thread before it ever reaches the AI. It’s enforced in code by a type-level gate and a build-time dependency rule — not just policy — so unredacted content cannot reach the model.
Drafts only — never auto-sent
C2Stream writes drafts for your review. Nothing is ever sent on a timer or on its own; every send is an explicit click after you’ve read the draft, and there’s a brief undo window before it goes out.
Your data never trains AI
We use Anthropic’s API, which does not train its models on submitted content, and we never permit your email content to be used to develop or improve generalized AI models. We extract your writing style and the thread’s meaning — for you, not for a training set.
Minimal mailbox access
Connecting Gmail grants only read access (to understand the thread) and compose access (to save and send your approved reply). We never request permission to modify or delete your mail.
Immutable audit trail
Every draft generated, saved, sent, or rejected is recorded to an append-only log enforced at the database level — entries can’t be edited or deleted, and each is integrity-hashed. It’s your record that the AI never acted without your review.
Built-in ethics tooling
Record informed client consent for AI use (ABA Formal Opinion 512) with an auditable timestamp, and see exactly what (redacted) content left your firm for each draft — turning confidentiality into something you can verify, not just trust.
Sub-processors
The third parties that process data on our behalf, and exactly what each one sees. The AI model receives only redacted text.
| Provider | Purpose | What it processes |
|---|---|---|
| Anthropic | AI drafting model | Redacted thread text only — never raw PII, and never used to train models. |
| Google (Gmail API) | Mailbox access | Read + compose on the mailbox you connect; we never request modify or delete. |
| Supabase | Database & authentication | Your account, settings, audit log, and encrypted mailbox tokens. |
| Vercel | Application hosting | Serves the app; processes requests in transit. No mailbox content at rest. |
| Stripe | Billing & payments | Your billing details. Card data is held by Stripe, never by C2Stream. |
| Resend | Product & account email | Your email address, to send account and product messages you can opt out of. |
Your data, your control
- No training, ever. Your email content is never used to train AI models — not ours, not a provider’s.
- Minimal retention. We don’t store your raw mailbox. Drafts and settings live in your account; the audit log records actions (who did what, when) — not message bodies, drafts, or tokens.
- Disconnect anytime. Disconnecting Gmail revokes our access immediately; you can also revoke it from your Google Account.
- Export & deletion. Request a copy or deletion of your data at any time — email support@c2stream.com and we’ll action it. The append-only audit log is retained for its integrity guarantee and contains no client content.
For the full legal detail, see our Privacy Policy and Terms. Reviewing C2Stream for your firm and need a DPA or security questionnaire completed? Email support@c2stream.com.