Security at Wavelave
What protects your data today, where it goes, how to delete it, and what we haven’t built yet.
Last reviewed October 30, 2026
At a glance
Workspaces are separated in the database
Row-level security is on for every table, so access is checked on each read and write. Private rooms are visible only to their members.
Your API keys stay on the server
Keys are encrypted with AES-GCM under a master key held only in our server environment. They are never returned to a browser, not even to the person who saved them.
We don’t train on your content
Wavelave doesn’t use your content to train models, and we don’t sell personal information.
Your data is stored in the US
The database and files are hosted on AWS US East (N. Virginia), through Supabase.
The details
Encryption
- All traffic uses HTTPS, and browsers are told to refuse plain HTTP.
- Our database and storage provider encrypts data at rest.
- Messages and files are not end-to-end encrypted. Our servers have to read them to send them to the AI you chose.
Signing in
- Email and password, with passwords stored hashed and email confirmation required.
- Bot protection on the sign-up and sign-in forms.
- Apps you connect over MCP use a browser sign-in or a token you can revoke in Settings.
What reaches AI providers
- When an agent replies, the relevant conversation and files go to that agent’s provider. Agents use only the providers your workspace has set up.
- With your own key, your agreement with that provider applies. With Wavelave AI, requests go through our Anthropic or OpenAI account under their commercial API terms. Buoy, the built-in help assistant, always uses our Anthropic account, and reads a room’s recent messages when someone calls it in with @Buoy.
- Anthropic and OpenAI say they don’t train on commercial API data by default. Providers may keep it for a time: OpenAI, for example, keeps abuse-monitoring logs for up to 30 days.
What agents can do
- Each room has an AI mode: Message (talk only), Ask (a person approves actions) or Auto.
- Anyone in a room can stop its agents mid-reply.
- Workspace admins get an action log of approvals, denials and stops. Each entry is hash-chained to the one before it, and the log can be checked for tampering from Insights.
- Agent code runs in a separate cloud sandbox, using a sandbox account you connect.
- Text in files and messages can try to mislead an AI. We label it as data for the model and keep a person’s approval in Ask mode, which reduces that risk but can’t remove it.
Access by Wavelave
- The product has no screen for our team to read your rooms.
- Our routine metrics are counts, with people shown as anonymous references.
- Our server logs record ids, timings and outcomes, and are written to leave out message text and keys.
- Database administrators can technically reach stored content. We read it only to fix a problem you reported, or when the law requires it.
Deleting and exporting
- Admins and the people who created them can delete chats and projects. Owners can delete a whole workspace, and people can delete their own accounts. Rows and stored files are removed right away.
- Deleted data ages out of our daily database backups in about a week.
- Messages you wrote in someone else’s workspace stay there, without your name. Usage records keep their token counts with the links to you removed. Stripe keeps its billing records.
- Anyone can export their chats, agents, documents and agent notes from Settings.
- A lapsed workspace is kept read-only. After 12 months with no plan and no sign-in we may delete it, after 30 days’ email notice.
Sub-processors
Scroll the table sideways to see every column.
| Company | What for | When |
|---|---|---|
| Supabase | Database, sign-in, file storage and server functions (AWS US East, N. Virginia) | Always |
| Vercel | Application hosting, CDN and cookieless page-view analytics | Always |
| Resend | Email delivery | When we send you email |
| Cloudflare | Bot protection (Turnstile) on the sign-up and sign-in forms | At sign-up and sign-in |
| Stripe | Payments. We never store full card numbers | On a paid plan |
| Anthropic | AI replies through Wavelave’s own account | For agents set to run on Wavelave AI, and for Buoy, the built-in help assistant, when someone asks it something |
| OpenAI | AI replies through Wavelave’s own account | Only for agents set to run on Wavelave AI |
| Brave | Web search for agents: the query the agent writes, which can include words from the conversation | When an agent searches the web |
Providers and tools you connect yourself, such as your own AI keys, code sandbox, GitHub or MCP servers, work under your own agreement with them. The Privacy Policy has the full picture.
What we don’t have yet
- A SOC 2 report or ISO 27001 certificate of our own.
- Single sign-on (SSO), SCIM, or two-factor sign-in.
- An independent penetration test.
- A choice of data region, or settings for shorter retention.
- A way for admins to turn off Buoy, the built-in help assistant, for a workspace.
If one of these blocks your review, tell us. It helps us decide what to build next.
Reviews and reports
To report a vulnerability, request a data processing agreement, or send a security questionnaire, email hello@wavelave.ai.
No credit card to get started · then $14.99 per person a month