What we actually do to protect your data

• Encryption in transit on every request (TLS 1.2+).
• Encryption at rest for the primary database and object storage.
• Row-Level Security in Postgres scopes every read and write to the workspace and user that owns the data. Cross-tenant reads are not possible through the API.
• Least-privilege roles. Application code uses a publishable key bound to the signed-in user; a service-role key is used only by server-side functions that need to bypass RLS, and never reaches the browser.
• Backups are retained by our database provider for point-in-time recovery.

When you generate content, your prompts and the relevant context (brand kit, knowledge documents, prior outputs) are sent to language-model providers to produce the result. We use providers under enterprise terms that contractually prohibit training on our customers' inputs or outputs. Current providers include Anthropic, OpenAI, and Google (Gemini). We also use ElevenLabs for voice transcription where you opt in to dictation, and Firecrawl/Browserless for fetching public web pages we research on your behalf.

We do not train our own foundation models on your data, and we do not share your content with third parties for advertising or analytics enrichment.

• Hosting & database — Supabase (Postgres, Storage, Auth) on AWS.
• Edge runtime — Cloudflare Workers.
• Model providers — Anthropic, OpenAI, Google Gemini.
• Web research — Firecrawl, Browserless.
• Voice — ElevenLabs (opt-in dictation only).
• Payments — Paddle (Merchant of Record; we never see full card numbers).
• Email — Resend for transactional and account email.

We'll publish material changes to this list before they take effect. Reach out via the in-app support channel if you need a formal subprocessor notification.

By default, our engineers cannot view the content of your workspace — we see operational metrics (error rates, latency, token counts, which step failed) with content redacted before it reaches our logs.

Two situations are exceptions, both auditable:

• Content review (opt-in). Workspace admins can turn on content review in Settings → Trust & Access. This lets our team look at prompts and generated outputs to spot quality issues and ship fixes. Off by default. You can turn it off at any time.
• Break-glass support. For severe incidents (data loss, hung jobs, account recovery) a small number of platform admins can access without prior consent. Every access is written to your workspace audit log with actor, timestamp, and reason — and workspace admins can read that log.

If we discover a security incident affecting your data, we will notify affected workspace owners without undue delay (and within the windows required by GDPR / UK GDPR). Notification includes what happened, what data was involved, what we've done, and what you can do.

We don't currently hold SOC 2 or ISO 27001. We won't claim certifications we haven't earned. SOC 2 Type II is on our roadmap; we'll update this page when we have engaged an auditor and started the observation window.

For enterprise procurement: we're happy to complete security questionnaires, sign a DPA, and provide a written statement of controls. Contact us through the app.

If you believe you've found a security issue, please report it via the in-app support channel with as much detail as you can share (steps to reproduce, affected URL, screenshots). We'll acknowledge within two business days. We don't currently run a paid bounty programme, but we credit researchers who report in good faith.