Vidya Terms of Service

By creating an account or using Vidya (the “Service”), you agree to these Terms. If you are using the Service on behalf of an organization, you confirm you have authority to bind that organization to these Terms.

If you do not agree, do not use the Service.

Vidya is an AI-driven data analysis and research assistant built for users who need enterprise-grade accuracy without enterprise-grade surveillance.

You can upload a spreadsheet or ask an open-ended question. The Service plans an approach, writes and executes code to answer it, and returns a written analysis with charts. It also supports document-based question answering, live web research, and custom specialized assistants (“subagents”) that you configure.

The Service uses a third-party large language model provider (currently Anthropic) to generate plans, code, and written responses. Section 4 explains exactly what that provider sees — and what it never sees.

You need an account, authenticated through our identity provider (Clerk), to use most of the Service. You are responsible for activity under your account and for keeping your credentials secure.

You must be legally able to enter into these Terms where you live. You may not use the Service if we have previously banned you from it, under any account.

This section is the one we’re most careful to keep accurate, because it’s a factual claim about running code, not a policy statement. Each row below reflects the actual, current retention behavior of the Service.

“Application database” rows are the exceptions; everything else about your uploaded content is designed to never reach durable storage at all.

This is where Vidya differs from consumer AI tools. These are not aspirational commitments — they are architectural constraints.

• We do not use your uploaded data, documents, or chat content to train any AI model — ours or a third party’s.
• We do not store your uploaded datasets or documents. They exist in server memory for the duration of your session (or up to one hour of inactivity, whichever is shorter) and are never written to a database.
• Answering your question is two separate AI steps with two different rules — read both points below together, not in isolation.
• Step 1, writing the code: personal details are never shown. Before any preview of your dataset is shown to the model that writes the analysis code — which needs to see column names and types, not real values — names, birthdates, government ID numbers, emails, phone numbers, and physical addresses are automatically detected and replaced with placeholders. This step never sees your real personal data at all.
• Step 2, writing your answer: this is a different, later step, and it works from your real data. Once the code has run, a second step writes your actual answer from the real, unredacted result — because that is the only way to correctly answer your question. If your question asks for identifying information, the answer will contain it: "who are my top customers" will name them. That is the correct answer, not a leak, and it becomes part of your kept chat history (Section 4). Regardless of what was asked, we always strip Social Security numbers, credit card numbers, and IBANs at this step before the model sees them or the answer is saved. Other categories — names, emails, phone numbers, physical addresses — are stripped at Step 1 above but are NOT currently stripped here, so they can and will appear in your final answer when they are genuinely part of the result.
• Requests are processed by our third-party model provider (currently Anthropic) to generate a response. Their handling of that data is governed by their own API terms. For Anthropic’s commercial API, inputs and outputs are not used to train their models by default (Is my data used for model training? — Anthropic Privacy Center). We encourage you to review their current policies directly, since third-party terms can change independently of this document.

Why this matters: Most AI tools centralize your data to improve their models and reduce infrastructure costs. Vidya does not. Your uploaded datasets and documents stay yours, processed in memory, and discarded when you’re done — the one deliberate exception is your own chat history (Section 4), which includes the answers written about that data. This is not a feature we added — it is how the system is built.

You may not use Vidya to:

• Attempt to bypass, disable, or manipulate its safety systems through adversarial prompts, prompt injection, or repeated reformulation of a declined request.
• Generate content intended to harass, defraud, or endanger any person.
• Upload data you don’t have the right to upload, or use the Service to violate someone else’s privacy.
• Probe, scan, or attempt to extract the Service’s internal architecture, prompts, or source code.
• Use the Service in any way that violates applicable law.

Every request is independently evaluated by an automated safety system before it’s carried out. That system is not perfect in either direction — see Section 7 for what happens when it declines a request, and how to contest a decision you believe was wrong.

We track, per account, how many times a request has been declined by our safety systems (Section 4). This count exists to catch a pattern of misuse — not to penalize a single ambiguous request.

What happens as that count rises

1. Automated review flag. Past a threshold, an account is surfaced internally for human review. Nothing happens to your account automatically at this stage.
2. Warning. If a reviewer determines the pattern reflects genuine attempts to bypass the Service’s safety systems, you’ll receive an email warning, with a way to reply if you believe it’s a mistake.
3. Suspension and removal. If the behavior continues after a warning, or is severe enough to skip the warning stage entirely (for example, content that could endanger someone), we may permanently suspend your account, delete your Service data, and cancel any active subscription.

Every step past the automated flag is a human decision, made by us, not a fully automated system. If you believe your account was actioned in error, contact us (Section 12).

Paid plans are billed on a recurring basis through our payment processor. Overage usage beyond a plan’s included quota is billed at the rate shown at the time of purchase. You can cancel at any time; cancellation takes effect at the end of the current billing period unless stated otherwise.

If your account is suspended for a Terms violation, any active subscription is cancelled as part of that process — you will not continue to be charged for a service you can no longer access.

You retain all rights to the data you upload and the questions you ask. You grant us only the limited, temporary right to process that content as needed to operate the Service and return a response to you — nothing more, and nothing that survives beyond the retention behavior described in Section 4.

This is not a “we own everything you upload” clause. You own your data. We process it to answer your question, then discard it.

The Service is provided “as is.” AI-generated analysis can be wrong, incomplete, or based on a misreading of your data — review anything consequential before acting on it. We do not warrant that the Service will be uninterrupted, error-free, or fit for a particular purpose.

We may update these Terms as the Service changes. If a change is material, we’ll make a reasonable effort to notify you before it takes effect. Continuing to use the Service after a change takes effect means you accept the update.

Questions about these Terms or how your data is handled: jw07@navi-os.cc