Data processing
What we process on your behalf, where it lives, and who else touches it — the annexes of a DPA, published so you do not have to ask for them.
Roles
For documents your people write in Specio, you are the controller and we are the processor. We process them only to provide the service and on your instructions. We do not sell them, and we do not use them to train models.
What we process
- Account data — name, email address, and the workspaces and organisations someone belongs to.
- Content — documents, comments, uploaded files and images.
- Learning records — for Specio Learn: enrolments, progress, answers and submissions, grades and feedback, certificates.
- Usage records — counts of AI actions, storage totals, and the days a member was active. Counts, never what was written.
- Administrative log — who invited whom, who changed a role or a setting. Deliberately contains no reference to any document.
Sub-processors
These are the third parties that may process customer data. We give notice before adding one. The list is also on the privacy policy.
- Supabase — Database, sign-in, file storage and live collaboration. Data: Account data, documents, comments, uploaded files, learning records.
- Vercel — Application hosting and edge delivery; cookieless Web Analytics and Speed Insights. Data: Every request passes through it (IP address, browser, page). Analytics are aggregate page views and performance timings.
- Resend — Transactional email (sign-in codes, invitations, notifications). Data: Email addresses and message contents.
- Inngest — Background jobs (long AI tasks, scheduled notifications, imports). Data: Job inputs, which can include document text and identifiers.
- Lemon Squeezy — Payments, as merchant of record for subscriptions and course purchases. Data: Name, email, billing address and payment details, which go to Lemon Squeezy directly; we receive order and subscription status.
- Railway — LaTeX compilation servers. Data: The LaTeX source and assets of the project being compiled.
- RunPod — GPU hosting for text recognition (OCR) and for models we run ourselves. Data: Images and PDF pages sent for text recognition; prompts sent to self-hosted models.
- AI model providers — OpenAI, Anthropic, Google (Gemini, Speech-to-Text), OpenRouter — Generating, reviewing and transcribing text when you use an AI feature. Which provider serves a request is configured by us per model. Data: The prompt, the parts of the document you include, and audio you dictate. Only when you use an AI feature. On Institution plans with your own provider credentials, prompts go to your provider under your contract instead.
- Langfuse — Tracing AI requests to debug failures and check quality. Data: AI prompts, outputs, timings and the account identifier. When enabled by the operator.
- Sentry — Error monitoring. Data: Error reports with the page, browser, IP address and the signed-in account identifier.
- Axiom — Application logs. Data: Request logs (path, status, timing, IP address). When enabled by the operator.
- PostHog — Product analytics. Data: Pages viewed, features used, device and browser. Only after a visitor allows analytics in the cookie banner.
- Umami — Cookieless page-view counts. Data: Page, referrer, browser and country; no cookie and no cross-site identifier.
Last updated 13 November 2026 (draft). The list follows what the code calls; it is a draft until counsel and the operator confirm it against deployed contracts and add regions.
Deletion and return
Any account can export everything it holds at any time, without asking us. Closing an account is scheduled thirty days out and reversible throughout. An organisation may set a retention period; it is off by default.
Security measures
Set out on the security page, including the structural separation that stops an organisation administrator reading a member’s documents, and our incident-notification commitment.
Getting a signed DPA
Write to legal@specio.space with your institution’s name and, if you have one, your own template — most universities do, and we would rather start from yours than argue about ours.
We deliberately do not offer a click-to-accept DPA. A data processing agreement is a contract between two named parties, and one signed against text nobody reviewed protects neither of us.