Offload privacy
Offload is designed around isolated accounts. One customer cannot use the application to read another customer's events, goals, preferences, receipts, billing state, or browser sign-ins.
When you choose “More like this” or “Fewer,” Offload stores bounded importance weights plus a tenant-specific keyed hash of the normalized sender and a small affinity score. You can also set those weights directly and provide reply-style guidance. Reply-style guidance is encrypted in storage and is supplied to managed reasoning only when Offload prepares an editable reply draft. The sender is not stored in plaintext in the preference table, and negative feedback alone cannot suppress deadlines, direct requests, academic or career obligations, or account and security alerts.
What is stored
We store account identity, verified email or Google identity linkage, a one-way salted password credential when you choose password sign-in, revocable session records, subscription status, usage counts, goals, preference weights, encrypted reply-style guidance, connector status, action receipts, approved outbound-message state, and information you direct Offload to ingest. Offload never stores your plaintext password or a reusable password-reset link. Passwords are processed only to derive or verify a versioned PBKDF2-SHA-256 credential; reset tokens are single-use, expire after 30 minutes, and are stored only as keyed hashes. Sensitive event text, Google credentials, approved reply text, provider send receipts, agent output, and reusable browser state are encrypted before database storage. Uploaded originals are not retained after bounded text extraction.
Service providers
Cloudflare provides the control plane, database, scheduled processing, and the managed AI inference used for customer workspace agents, natural-language group interpretation, and declarative view adaptation. Launching an agent sends the instruction, selected starting objects, and current tenant workspace context assembled server-side from connected events, active goals, reusable groups, and agent-created objects. The enabled-by-default background system may also review current tenant-owned workspace context, propose a reusable subgraph, and prepare a declarative object-layout preview once per day when graph evidence changes; you can inspect, run, pause, or resume it from the agent panel. General agents may create artifacts and update Offload's internal graph with retained provenance. Consequential connector actions are handled by Offload's action-specific autonomy policies and receipts; connector credentials are not placed in model prompts. Generated subgraphs require your confirmation and generated layouts remain previews until you keep one. A technically isolated owner-only prelaunch experiment may use a separate server-side model route, but that route is never selected for customer tenants. Vercel hosts the web application and server-side Stagehand runtime. Browserbase provides isolated cloud browser sessions, Live View, and persistent site contexts for websites you explicitly connect. Stripe processes subscription payments; Offload stores Stripe identifiers and status, not full card details. Google processes identity, Gmail, and Calendar authorization when you connect it. Resend processes operational email delivery to your verified account address when notifications are enabled. Other connected websites process information under their own terms when you choose to use them.
Google data
Offload requests read-only Gmail access, send-only Gmail access, and Google Calendar event access to surface obligations and deadlines, send only the final replies you explicitly approve, and create private holds only after your explicit action review without attendees or invitation updates. Gmail synchronization starts with a bounded recent window, then uses Google's incremental history stream and push notifications when available. Offload extracts bounded message text, does not ingest attachment contents, and encrypts stored message text, approved replies, provider receipts, and Google credentials separately for each tenant.
Offload's use and transfer of information received from Google APIs adheres to the Google API Services User Data Policy, including its Limited Use requirements. Google user data is processed only to provide the user-facing productivity features described here. Offload does not sell it, use it for advertising, use it to train generalized AI or machine-learning models, or let another customer access it. Offload operators do not read message content except with your affirmative permission for specific support, when necessary to investigate abuse or a security incident, or when legally required. Every send requires a direct user action on the final draft. Offload refuses automated and bulk reply targets, records a durable encrypted receipt, deduplicates by source event, and does not retry an ambiguous provider submission automatically. Disconnecting Google deletes the local credential and imported Gmail and Calendar events, removes linked send, notification, and agent-run records, stops future synchronization, and requests upstream revocation.
Managed reasoning
Your subscription includes tenant-isolated managed reasoning. Context-agent and group outputs are encrypted at rest and remain proposed until you accept an artifact or approve a graph update. They cannot silently send, publish, purchase, schedule, authenticate, or mutate an external system. Offload does not use your content to train a generalized model. Managed reasoning does not share or consume another person's ChatGPT credentials, and the operator's optional owner-only model route remains technically and financially separate from customer accounts.
Proactive notifications
Email notifications are enabled by default for approaching deadlines and completed scheduled goal reviews. Offload derives the recipient from your verified account instead of accepting an arbitrary address. From the account page, you can choose immediate or daily-digest delivery, set quiet hours and a daily interruption budget in your selected time zone, or turn proactive email off. Alerts deferred by quiet hours or the daily budget are bundled into the next digest instead of dropped; disabling email suppresses queued notifications. Provider receipt identifiers are encrypted before storage, and completed or suppressed delivery rows are removed after 90 days.
Control and retention
You can download a portable JSON Lines archive of your graph objects, groups, agent artifacts and proposals, preferences, source history, prepared actions, and audit records from the account screen. The archive excludes password and recovery material, active login sessions, OAuth credentials, browser cookies and provider session handles, service credentials, notification-provider receipts, internal operational incidents, and Stripe identifiers.
You can reset or add a password through a verified one-time email link, disconnect Google, remove an individual reusable browser session, turn proactive email off, cancel billing in the Stripe customer portal, sign out, and permanently delete your account and tenant data from the account screen. Resetting a password revokes every older Offload session. Account deletion deletes the password credential, pending recovery links, reusable connector credentials, browser state, events, goals, notification state, outbound-message state, preferences, receipts, agent output, and active sessions. Minimal payment or security records may be retained by service providers where necessary for fraud prevention, billing disputes, and legal compliance.