Legal
Privacy Policy
What Specula stores about you, why, which services process it on Specula's behalf, how long it is kept, and how to see, export or delete it.
Effective 19 October 2026
In short
- Your conversations, reports, boards, memories and uploads are stored so you can come back to them. They are private to you unless you share a link, or choose to share one question and answer with Specula's reviewers when you give feedback.
- To answer you, your messages and uploads are processed by the AI and research services Specula is built on, listed below.
- Specula does not use your conversations or uploads to train AI models unless you turn that on for your reviewed feedback (off by default), does not sell your data, and has no advertising or analytics tracking.
- AI output can be wrong. Don't submit passwords, payment details or other highly sensitive information.
- You can export your data, delete conversations, manage memory, or delete your account at any time in Settings.
1.Who is responsible for your data
Specula is operated by an individual, not a registered company. The operator's name and contact address have not been published in this draft yet; they will be added here before Specula's public launch.
The operator decides how Specula processes personal data and is responsible for it (the “controller”).
2.What Specula stores
| What | Details |
|---|---|
| Account (Google sign-in) | Your name, email address and Google profile picture, the identifier and sign-in tokens Google issues for your account, and your sign-in sessions (including the IP address and browser they were created from). Specula only asks Google for your basic profile. |
| Age confirmation | Your date of birth, used only to confirm you are 18 or over. As a guest it is kept in a signed cookie on your device; with an account, on your profile. |
| Profile and settings | What you choose to tell Specula: role, interests, preferred answer style and language, standing instructions, theme, and whether memory is on. |
| Conversations and research | Your messages and Specula's answers in full, the research steps and sources behind them, and the reports, comparisons and learning modules created in them (with your edits and practice progress). |
| Memory | Facts you ask Specula to remember (when memory is on). Specula only saves, changes or deletes a memory on your own words, never because a web page, file or tool result asked it to. |
| Files and images | Files and images you attach to research messages, your profile picture, screenshots you add to Xavi boards, and images and videos you upload to Vision, with Vision's analysis results. |
| Xavi | Your boards and their history, snapshots, problem progress, and your latest Xavi conversation (the most recent 200 messages). |
| Sharing | Share links you create, with their settings (a securely hashed password if you set one, expiry, view count and when they were last opened). |
| Usage and safety | Which features you used and when (for allowances), a random guest identifier on your device, when you were last active, account restrictions if any, and an audit log of administrator actions. For guests, messages and agreements to the Terms are also counted per network using a one-way hash of the IP address (never the address itself), so the free allowance can't be renewed by clearing cookies. |
| Terms acceptance | When you agreed to the Terms of Use and confirmed the Privacy Policy, and which version. |
| Valuation records | Each transfer valuation Specula makes is logged with its inputs, so its accuracy can be checked against real fees later. |
| Feedback | When you rate an answer, a report, a recommended player or which of two answers was better: your rating, the reasons and note you add, which output it is about, the AI model and version that produced it, and whether you shared that question and answer with reviewers. Reviewers' judgements of it. If you describe your work for reviewers, that description, your expertise and any conflicts of interest you declare. Whether you allow your reviewed feedback to be used to improve Specula's models, and when you said so. |
As a guest, your research is stored against the random guest identifier on your device. If you sign in on that device, it moves to your account; otherwise it is deleted after 30 days without use (see section 6).
3.Why Specula uses it
- To provide the service you asked for — research, answers, reports, Xavi, Vision, saving your work, memory and sharing. This is necessary to provide Specula to you under the Terms of Use.
- To keep Specula safe and fair — sign-in security, age checks, usage allowances, preventing abuse and automated misuse, and checking content for safety. These are Specula's legitimate interests in running a secure service.
- To check the valuation model — comparing logged valuations with real transfer fees. This is a legitimate interest in keeping valuations honest; the records are anonymised when you delete your account.
- To evaluate and improve answer quality from your feedback — reviewers check the feedback you choose to give (and the question and answer only if you shared them with it), to find and fix recurring mistakes and to measure models against each other. This is a legitimate interest in giving accurate answers; giving feedback is always optional.
- To train or evaluate Specula's models with your reviewed feedback — only if you turn this on in Settings → Privacy & data (it is off by default). This relies on your consent, which you can withdraw at any time.
Apart from that optional switch, Specula does not rely on consent for these, and does not use your data for advertising, profiling for marketing, or selling to anyone.
4.AI processing and service providers
Specula is built on the following services, which process data on its behalf to provide the features below:
| Provider | What it does for Specula | What it receives |
|---|---|---|
| Vercel | Hosting the website and storing files (private storage) | Everything sent to and from Specula; your files and images |
| Neon | The database | Everything listed in section 2 |
| Anthropic (United States) | Research's AI model (Claude): answers, research reasoning and reports | Your Research messages and conversation context, your name and the profile facts relevant to the answer, the memories saved in Research (never Xavi's or Vision's, nor your Xavi conversations), attached files and images (through short-lived private links), and a pseudonymous ID for your account (never your email) |
| OpenAI | The other AI models: understanding each Research request, page reading, Research answers when Claude is unavailable (only if that backup is turned on), Xavi's reasoning and voice, Vision answers and image analysis, and content safety checks | Your messages, conversation context, memories and profile facts relevant to the answer, attached files and images, and Xavi voice audio (sent from your browser directly to OpenAI while you talk to Xavi) |
| Sign-in | Your sign-in request | |
| Exa and Firecrawl | Web search and reading web pages | Search queries and page addresses Specula chooses while researching your question |
| StrikerAPI (United States) | Football statistics, when enabled | Player, club and competition names and ids to look up — never your messages or account details |
| TypeSafe | Short judgements about Xavi boards, when enabled | A text description of the board |
| Modal | Vision video analysis, when enabled | The video you uploaded, through a short-lived private link |
Specula also looks up public information on YouTube (to check videos), Wikipedia and Wikimedia Commons (player photos). Your browser loads some images directly from those services and site icons from Google, which means they see your IP address; Specula sets no cookies for them. YouTube videos and Sofascore widgets load only when you click them.
Training. Specula does not use your conversations, memories or uploads to train or fine-tune AI models. The one exception is your choice: with “Use my reviewed feedback to improve Specula's models” turned on in Settings, feedback you give that reviewers accept, and the questions and answers you shared with it, may be included in datasets used to train or evaluate Specula's models, with your name and contact details left out and personal details removed. Uploads and Vision media are never included. Turning it off, retracting feedback or deleting the conversation takes it out of every later dataset; a model already trained on it can't unlearn it, but it is left out when the model is next trained. How each provider may use the data it processes for Specula is governed by its own terms for business customers.
Untrusted content. Web pages, search results, posts and files can contain text written to manipulate an AI. Specula treats them as information, never as instructions: it does not let them steer your private details out of Specula or change your memory.
5.Where data is processed
These providers may process data in countries other than yours, including the United States. Data is sent to them over encrypted connections.
6.How long data is kept
| Data | Kept until |
|---|---|
| Account, profile and settings | You delete your account |
| Conversations, reports and their files | You delete the conversation, use Delete all research, or delete your account |
| Memories | You delete them, clear memory, or delete your account. A deleted memory stops being used immediately and is kept hidden for up to 30 days so open conversations can be told to disregard it; account deletion removes it at once. |
| Xavi boards and snapshots | You delete them or your account. Only the most recent 200 Xavi messages are kept. |
| Vision uploads and results | You delete the Vision conversation, use Delete all research, or delete your account. Temporary files used during an analysis are deleted when it ends, except that a full match's video is also cached on the analysis service's temporary disk while it is analysed and for a while after: deleting the match asks the service to remove that copy, and the copy also goes when that temporary disk is cleared. If the service doesn't confirm the removal, Specula repeats the request for at least a week, or for up to 30 days while the service can't be reached; a removal still unconfirmed after that is flagged to Specula's operators to remove by hand. |
| Sign-in sessions | 30 days, or until you sign out |
| Share links | You revoke them, they expire, or you delete what they share |
| Usage records | With an account, until you delete your account. As a guest (including the per-network counts), deleted 30 days after the allowance they counted towards has reset. |
| Terms acceptance | With an account, until you delete your account. As a guest, moved to your account if you sign in on that device; otherwise deleted 395 days after you agreed, once your guest research has been deleted. |
| Guest research | Moved to your account if you sign in on the same device. Otherwise each conversation is deleted once it hasn't been used for 30 days, with everything stored for it. |
| Valuation records | Kept for checking the model; your account link is removed when you delete your account |
| Feedback | You remove it, delete the conversation it is about (or all research), or delete your account. A reviewer's judgement stays with the reviewer's work after the reviewer's account is deleted, without their identity. |
| Administrator audit log | Kept for accountability, including after an account is deleted |
7.Your choices and rights
- See and export a copy of your data — profile and settings, memories, conversations with their messages (and the ones you withdrew), reports, learning progress, your Xavi notebook, the links you created (without their passwords), your agreements to the Terms, and a list of your files and Vision uploads — as a file: Settings → Privacy & data → Export your data.
- Correct your profile in Settings, and your memories in Settings → Memory (or ask Specula in your own words).
- Delete a conversation from History, all research in Settings → Privacy & data, memories in Settings → Memory, or your whole account in Settings → Account → Delete account.
- Turn memory off in Settings → Memory.
- Remove feedback you gave from the answer it is about, and turn model training on or off for your reviewed feedback in Settings → Privacy & data.
Depending on where you live, you may also have the right to access, correct, delete, restrict or object to the processing of your data, to receive it in a portable format, and to complain to your data protection authority. Contact the operator to use any of these rights.
8.Security
Files are kept in private storage and opened only through short-lived signed links after Specula checks you may see them. Connections are encrypted, access is checked on the server for every request, and secrets never reach your browser or the AI. No online service can promise perfect security; if a breach affected your data, we would tell you as the law requires.
9.Children
Specula is only for adults aged 18 and over. People who say they are under 18 are not let in.
10.Cookies
Specula uses only cookies and browser storage that it needs to work, plus preferences you set. The Cookie Policy lists them.
11.Changes
When this policy changes in a way that matters, Specula shows the new version and asks you to confirm it before your next message.
12.Contact
The contact address for questions and data requests will be published here before Specula's public launch. Until then, signed-in users can export or delete their data themselves in Settings.