Trust & Security
Aflow is an agent platform: you build and run AI agents that work with your data, your model keys, and the services you connect. That only works if you can trust exactly where your data goes. This page explains it in plain language.
Where your data lives
Your workspaces, conversations, memories, files, and run history are stored on Google Cloud in Frankfurt, Germany (region europe-west3) — database, file storage, and agent execution all run there. The web application is delivered via Vercel, and sign-in is handled by Auth0 on an EU tenant.
Who can see your content
Content is scoped to the workspace it lives in. Personal workspaces are visible only to you and members you explicitly add. Shared workspaces are visible to their members according to each member’s role — collaborators can see that workspace’s chats, runs, memories, files, and connected resources. Platform administrators hold management visibility only (workspace lists, membership, quotas, error rates, abuse signals) and are technically excluded from reading the content of a personal workspace they are not a member of. That exclusion is enforced in the authorization layer and covered by automated tests, not just policy. You decide whom you invite, and you are responsible for having the right to upload third-party data.
Your keys and connected accounts
- Model API keys and integration credentials are encrypted at rest and never displayed again after you enter them.
- Keys are sent only to their own provider (your OpenAI key goes to OpenAI, and nowhere else).
- Agents never see credential values — they reference credential slots; the platform injects the secret at the network boundary.
- OAuth connections (for example Google) request only the scopes the specific integration declares, and you approve them in the provider’s own consent screen.
What leaves the platform
Content leaves Aflow mainly toward services youconfigure: the AI providers you bring keys for, and the integrations you install and authorize. Two narrow, platform-run exceptions exist: the memory index uses the platform’s embedding provider to make your notes searchable (automatic and spend-bounded), and, if you start a voice conversation, your audio is streamed to the platform’s voice providers (transport, speech-to-text, and text-to-speech) for the duration of that call. Both are described in the privacy policy.
Agents act under an approval policy
Actions your agents take toward external services — sending, posting, ordering, modifying — are governed by the approval policy configured for the workspace. Medium- and high-risk writes require your confirmation in the app by default; low-risk or explicitly pre-authorized actions may run without a per-action prompt. Agent capabilities are bounded by a safety profile: agents in personal workspaces cannot execute arbitrary code on platform infrastructure or reach network hosts their integrations have not declared.
Telemetry
The web app collects cookieless, aggregate page analytics (no advertising trackers, no cross-site profiles). There are no analytics on your content — prompts, memories, and run outputs are never used for analytics or model training by Aflow.
Beta expectations
Aflow is an invite-only beta operated with care but without a formal SLA. The primary database is backed up daily. If something looks wrong, or you find a security issue, write to support@aflow.ai — security reports are triaged first.
Deleting your data
You can permanently purge any workspace yourself, including all of its content, from workspace settings. For full account deletion during the beta, email support@aflow.ai; we action deletion requests without undue delay and at the latest within one month, covering your account identity, personal workspaces, stored credentials, and connected accounts.