What we collect, why, and how we protect it.

agentplain is a service partnership. To do the work we promise, we need read access to your operating systems — email (Gmail or Microsoft 365), calendar, your file substrate (Google Drive or OneDrive), your accounting tool (QuickBooks Online), your CRM, and your transaction or document management system depending on vertical. You authorize each connection explicitly through the integration tile's OAuth flow.

Connection scopes are the minimum needed to deliver the value loop (read, categorize, coordinate, schedule, draft). We never request send-on-your-behalf scopes on email — the fleet drafts into your inbox; you send from your own account.

We also collect account-level data you give us directly: your name, business email, business name, billing details, the vertical you picked, and the configuration choices you make in the workspace (tone, default hours, scheduling preferences, skill selections).

Every piece of data we hold is scoped to a single customer workspace. The fleet uses it to do the work you hired us to do: categorize inbound, draft replies, propose meeting times, surface compliance flags, generate briefings. Each draft lands in your approvals queue. Nothing sends outbound on your behalf — your existing email, calendar, and CRM execute every send from your own account.

We use connected systems' data to train the fleet on your voice and preferences (which we store as an append-only feedback log in your workspace). We do not train any base model on your data, do not share your data with any AI model provider's training pipeline, and do not pool data across customers.

We never use your data to train models. Your chats with Plaino and the records your connectors expose are used to do your work and nothing else. We do not fine-tune or improve any model on them, and we do not feed them into a training feedback loop — our own or a provider's. Our model provider does not train on inputs or outputs of its commercial API by default (it is named in the subprocessor list below).

The only identifier we attach to a model request is a one-way scrambled hash of your workspace id. It lets the provider isolate a single abusive workspace if necessary, and it reveals nothing about who you are — never your name, email, or business name.

Separately, we keep an internal access-pattern log — which workspace ran a chat or read from a connector, and when — strictly for security and abuse review (see our acceptable use policy). That log records coarse metadata, never the contents of your chats or records, and it is never used for training.

Customer-facing payloads (approval queue items, handoff log entries, the documents you connect or upload) are encrypted at rest using AES-256-GCM with a per-environment key. Production secrets are stored only in our hosting provider's secrets store (Vercel) and never in source control.

Database rows carry a workspace_id column enforced by row-level security policies — queries are constrained at the database layer so a leaked query helper cannot return another customer's data. The same isolation extends to vector embeddings: every chunk of your uploaded and connected documents is workspace-scoped.

OAuth tokens for your connected systems live encrypted alongside the rest of your workspace state. Token refresh happens server-side; tokens never leave our infrastructure.

We use a small set of named subprocessors. Each is contractually bound by their own data-processing terms; we are responsible for our choice of them and for the configuration of our use.

• Anthropic — model inference. On the no-training API tier; your data is not used to train models. Customer data passes through model context only for the duration of an inference call.
• Neon — Postgres database hosting. Encrypted at rest + in transit; daily backups.
• Vercel — application hosting + edge network.
• Stripe — payment processing. We never see or store full card numbers; Stripe holds the payment method, we hold a customer ID + last-four.
• Resend — transactional email (sign-in links, billing receipts, support replies).
• Sentry — error monitoring. PII scrubbed at the edge before events leave our infrastructure.
• Inngest — event + cron orchestration. Receives event payloads but not OAuth tokens or full document bodies.

You can export your workspace data and close the workspace from inside the product at any time. Workspace closure triggers a 7-day soft-delete grace window during which you can restore (a walk-away under the trial guarantee deletes right away, with no grace window); after that, your workspace data is hard-deleted from our primary database. What stays after closure: your workspace’s name and settings, the reason you gave for closing (if you gave one), your team’s accounts (names and email addresses), your billing records, the record of when you agreed to these terms, and the record of what you approved or declined, and what was sent. We keep the last two because they are the legal proof of what you agreed to and what you authorized. Backups retain encrypted snapshots for an additional 30 days for disaster recovery; we do not restore from backups except to recover from a service-affecting incident, and we never read backup contents.

If you have a deletion request that's broader than the in-product controls, or any other data-subject question, email hello@agentplain.com and we'll handle it in writing.

Depending on where you and your clients are located, you may have statutory rights over personal information we process. We honor these regardless of jurisdiction:

• Access + portability. Get a copy of the data we hold for you — one tap on the Your data page produces a full structured export.
• Deletion / erasure. Delete your data. Closing your workspace purges your customer data after a short grace window; walking away under the trial guarantee deletes it right away, with no grace window. What stays after closure is listed in § 06.
• Rectification. Correct inaccurate information — edit it in the product, or ask us in writing.
• Restriction + objection. Ask us to pause or limit processing, including pausing the fleet entirely from your settings.
• No sale of personal information. We do not sell your or your clients' personal information, and we do not share it for cross-context behavioral advertising — there is nothing to opt out of, because we never do it.
• No automated decisions with legal effect. The fleet drafts and proposes; you review and decide. We do not make automated decisions about your clients that produce legal or similarly significant effects.

For data we process on your behalf, you are the data controller and we are your processor; we act on your documented instructions. To exercise any right, or to request a data-processing addendum, email hello@agentplain.com. We do not charge for reasonable requests and respond within the timeframes the applicable law requires.

agentplain is not a licensed broker, lender, carrier, attorney, CPA, RIA, or any other regulated party in the verticals we serve. Liability for licensed activities — broker-of-record decisions, tax filings, legal advice, fiduciary recommendations, insurance placements — stays with you and your firm. Every customer-facing output is drafted, queued for your review, and sent (if at all) by you from your own systems.

Material changes are announced by email to your workspace owner at least 30 days before they take effect. The latest version always lives at this URL.

Contact: hello@agentplain.com. Or read our terms and security pages.