Privacy Policy

Who this covers

This policy describes how ANSOFF AI handles information when you use the product. ANSOFF AI is operated by [Company legal entity], [Address]. For a privacy request, write to [privacy contact email]. You can also download your data or delete your account from Settings without waiting on that inbox.

What we collect, and why

• Account information. Your name, email address, and a password hash (or a Supabase sign-in id when that is turned on). We use it to create the account, sign you in, and send confirmation and password-reset messages.
• Workspace information. The firm name you type, and your role in that workspace. We use it to keep one firm's work separate from another's.
• Uploaded client files. The files you add to an engagement, the text extracted from them, and the search embeddings built from that text. We use them to answer questions, run the calculations, and produce the deliverables.
• Prompts and outputs. The questions you type and the answers, plans, and deliverables ANSOFF produces. We store them so you can come back to the work.
• Usage logs. Records such as uploads, downloads, deletions, invitations, and model token counts (the count, not the text). We use them to operate the service, investigate errors, and apply plan limits.
• Cookies. A sign-in cookie and a workspace cookie. Both are required for the product to work. We do not use advertising cookies or a third-party analytics cookie. See Cookies for the list.
• Signup source. If you arrive from a link with campaign parameters, we store those parameters and the referring site so we know which page you came from. That record lives with your account. It is not sold.

We do not sell personal information.

How AI is used

When an administrator has configured a model key, the passages needed to answer a question are sent to Anthropic's API to write the reply. If OpenAI is configured as the model or the fallback, those passages can be sent there instead. Extracted passages can also be sent to OpenAI to build search embeddings, and only when that key is set.

Web research sends the search query to Tavily. It does not send your uploaded files. Exa or Brave Search receive the query only when their keys are the ones configured.

Answers are generated by software. They can be wrong, incomplete, or out of date. They are not professional, legal, or financial advice. You have to check them before you rely on them or hand them to a client. The product says when a figure was calculated by code and when a source was not verified.

Your data is not used to train models. ANSOFF AI does not keep a training corpus, does not opt in to provider training, and does not write the text of your files, prompts, or outputs to logs that could be used for training. Usage rows store token counts. Error reports have personal data removed. Anthropic's and OpenAI's commercial API terms say API inputs and outputs are not used to train their models by default, and ANSOFF does not switch that on.

Companies that process data for us

These companies process data on our behalf. Optional ones run only when that integration is switched on.

• Supabase. Hosts the database, sign-in, and the private file buckets.
• Vercel. Hosts the application.
• Anthropic (only when configured). Writes answers from the passages and the question needed for that request, when an API key is configured. ANSOFF does not turn on training.
• Tavily (only when configured). Web search. ANSOFF sends the search query, not your uploaded files.
• OpenAI (only when configured). Builds search embeddings from extracted passages, and can write answers only if it is configured as the model or the fallback. ANSOFF does not turn on training.
• Exa (only when configured). Web search, only if that key is set instead of or in addition to Tavily. The search query is sent, not your files.
• Brave Search (only when configured). Web search, only if that key is set. The search query is sent, not your files.
• Stripe (only when configured). Takes payment when billing is turned on. ANSOFF is in test mode until a live key is deliberately enabled.
• Sentry (only when configured). Receives error reports when a DSN is set. Personal data is scrubbed and session replay is off.
• Upstash (only when configured). Stores rate-limit counters when those keys are set. It does not receive your files or prompts.

There is no third-party product analytics and no ad network.

How long we keep it, and how you delete it

Unfinished uploads are deleted after 24 hours. Active files and engagements stay until you delete them, or until you delete the account that owns them. After an account is deleted, any anonymized row that could not be removed immediately is purged within 30 days. Audit events are kept for 365 days. Sign-in sessions expire after 14 days, or sooner when you sign out.

You can delete one file, one engagement, or the whole account. Deleting a file removes the stored object, the extracted text, the embeddings, and the database rows. Deleting an engagement does the same for everything in it. Deleting an account removes every workspace you alone own, including those files. In a workspace that has another owner, your membership and your conversations are removed, and files already shared with that workspace stay with the firm.

Account deletion is not a hide-and-keep. The rows and storage objects are removed. If a row cannot be removed in that moment, it is anonymized and the daily job deletes it within 30 days.

Settings also has a download of your account record: profile, workspaces, engagements you own, file names you uploaded, and your prompts and answers.

Security

Workspaces are isolated with database row-level security. User files sit in a private storage bucket. Downloads are checked on every request. We do not claim SOC 2, ISO 27001, HIPAA, or any other certification.

Contact

Privacy requests: [privacy contact email]. Until that address is connected, use Settings to download your data or delete the account. [Company legal entity], [Address].