Storied AI: Data Privacy

That’s the account data that runs the service.

Separately, and only while a generation is running, we hold the request you sent and the result that came back. The next section explains exactly what that means.

The 30-Minute Window

This is the part most privacy pages skip, so we’ll be specific.

Some Storied skills run on your own device, and when they do, your content never leaves your machine. Others generate work in our cloud. For those, your request has to reach us. There’s no way to produce the artifact otherwise.

Here’s the sequence. You send a request. We write it to our database along with the job. The work is generated. We write the result to the same place. You collect it. An automated cleanup deletes the whole row, request and result together, 30 minutes after the job was created.

While that row exists, the only things that touch it are automated: the system producing your result, and the automated scoring described in the next section. We don’t read your job data, and outside the one narrow case described in “What We Can’t See,” we keep no copy of it anywhere else in our own systems.

Two qualifiers rather than a guarantee we can’t enforce. Our hosting provider keeps its own short-term database backups as part of running the infrastructure, the way every cloud database does, and those are outside our reach to delete row by row. And a request reaches Anthropic’s model to be answered, which is covered further down.

That 30 minutes isn’t a retention policy chosen for our convenience. It’s the shortest window the job can exist in and still hand you back your work.

How We Measure Quality (Beta)

Storied AI is in beta, and checking how the skills perform is part of what a beta is. Two things do that work, and both are narrower than you’d expect.

Quality scores, kept as numbers. Every generated result is checked automatically against a set of internal writing standards, things like whether it reads like a finished piece rather than a template. We keep the score and the counts. We don’t keep the sentences that produced them. This check runs on every job, and what it leaves behind is a number.

Which kind of job you ran. Every skill has a fixed category: investor pitch, strategy document, keynote, and so on. That category comes from the tool you invoked, not from anything reading your request. We don’t pass your writing through a model to classify it. The label was already in the request.

There is a deeper kind of review, which means holding actual requests and results long enough to study them and improve the skills. It is switched off on the service you use. The capability exists in our code and we run it inside our own environment, against our own team’s work. If we ever intend to turn it on for customer work, we’ll update this page and tell you before it happens.

We also run a design partner program, where some people agree in writing to share specific pieces of their work with us so we can improve the skills. That’s separate from this page, it’s something you opt into rather than something that comes with the service, and declining costs you nothing. Uploaded files are outside it. If you haven’t signed one of those agreements, the paragraph above is what applies to you.

Your uploaded files are never sent to us, for scoring, for categories, for the design partner program, or for anything else.

What We Can’t See

The design does the work here.

We don’t see your Claude conversation. You set up Storied AI by adding its MCP connector inside Claude, and the back-and-forth between you and Claude isn’t ours. What we receive is scoped to a job: the request you sent to a Storied skill, for the 30 minutes described above, and nothing surrounding it.

Your uploaded files aren’t sent to us on any plan or in any mode.

The activity log names the action, not the substance. It records that a skill ran, at a time, with an allowed or denied result. The names of the skills you run are part of that log, so the log knows what you did, not what was in it.

One narrow place is different, and it’s the exception to everything above. When a job fails, the entry carries a short technical description of the failure from our own system. Those descriptions can include a fragment of the request that caused them, so we truncate them and strip the payload before they’re written. We won’t claim it’s airtight, because no filter is. What we can tell you is that this is the only route by which your words could reach a record in our own systems that outlives the 30-minute window, and it’s built to keep them out.

And what we do keep, we keep clean. Sign-in secrets are never stored in a readable form. Your email is masked in our operational logs, and access tokens and secrets are stripped out before anything is written down.

Your Work Runs on Your Own Claude

Your work goes to the Claude you already run. Storied AI is an MCP connector inside Claude, not a second AI service that copies your conversation somewhere else. The model doing the reasoning is Claude, built by Anthropic. How your conversation is stored, retained, or used is governed by your own agreement with Anthropic and by your settings in Claude, not by us. We can’t see those conversations, and we won’t claim terms over them that aren’t ours to make.

Anthropic shows up twice, and both roles are worth naming. Claude is yours, running under your account. And when a Storied skill generates work in our cloud, we send that request to Anthropic’s model under our own account to produce the result and to score it. So for those jobs Anthropic is our supplier as well as your provider, and it’s covered by our agreement with them, not only yours.

Three practical notes so you can set this up with confidence:

• We recommend running Storied AI on a paid Claude plan. MCP connectors work on the free plan too, with a limit, but a paid plan is the smoother home for this kind of tool.

• Your Claude plan and data settings decide whether Anthropic can use your conversations to improve their models. On Claude Team and Enterprise plans that setting is off unless someone turns it on. On personal plans, it’s a control you own and can change in a minute. You’ll find it in Claude’s own privacy and data controls, and Anthropic’s own policy is the source of truth for what they do, not this page.

• A performance note, not a privacy one: Storied AI does its sharpest work on Claude’s most capable models. Set your Claude model to Opus, or Fable, when you run it.

Pick your Claude plan and settings deliberately, and you’ve set the terms for your own content. We’re the layer that checks your access, and we stay out of the conversation.

How Long We Keep It

Four kinds of record, four clocks.

1. Your account, meaning email, access level, approved tools, devices, and sessions, stays while your account is active.

2. Job data, meaning the request and the result of a generation, lasts 30 minutes, then an automated cleanup deletes it.

3. The activity log, meaning which skill ran, when, and allowed or denied, is kept for 30 days, then deleted automatically. We hold it that long for two reasons: it’s how we catch abuse of the service, and it’s the only record that lets us answer honestly if you ask us what we did with your data.

4. Run counts, meaning tokens, timings, model, and quality scores, are kept for 30 days, then deleted automatically on the same clock as the log. We used to describe cutting your identity off those rows and keeping the numbers. Deleting them outright is simpler and it’s stronger, so that’s what we do.

We do keep aggregate statistics that aren’t tied to any account, like how many runs happened in a month and how scores moved over a quarter. Those are totals, and there’s nothing in them to trace back to you.

Your Rights, and How to Use Them

Whoever you are and wherever you are, you can do all of this:

• See what we hold. Ask us and we’ll tell you.

• Correct it. If your account details are wrong, we’ll fix them.

• Delete it. Ask us to delete your account and we remove your email, your devices, your sessions, your approved tools, and your access records. The operational rows tied to you are deleted too, not stripped of your name and kept.

• Withdraw a tool approval. You approved each Storied skill Claude can call. You can take any of those back inside Claude, and you can disconnect the whole connector.

• Object. We keep the activity log and the run counts because we have a legitimate interest in a secure, working, improving service. If you’d rather we didn’t, write to us and say so, and we’ll weigh it and reply. This isn’t a form to fill in.

• Leave. Disconnect the MCP connector inside Claude anytime, which stops new activity immediately. Records already written age out on the clocks above. You can also revoke an individual device or sign-in session from your account.

We chose not to gate these by region. If you’re in the EU or the UK, this covers your GDPR rights including access, correction, erasure, portability, and objection, and you can also complain to your national data protection authority if you think we’ve got it wrong. If you’re in California, it covers your CCPA and CPRA rights, and the short version there is that we don’t sell your personal information and we don’t share it for cross-context behavioral advertising, so there’s no opt-out to chase.

To use any of these, email support@storiedinc.com.

The Rest

Who we are. Storied AI is operated by Get Storied, Inc., a Nevada corporation doing business as Storied. For privacy questions or deletion requests, contact us at support@storiedinc.com. Written notices can be sent to 611 Wilshire Blvd, Suite 900, PMB #977, Los Angeles, CA 90017.

The companies behind the service. We keep the list short on purpose. Here’s all of it.