Status: unreviewed draft. Not in effect. No lawyer has read this document. It is published so that it can be corrected, not so that it can be relied on. Where it describes what the system does, it was written against the running code rather than against the plan, and the sections that say a control is not yet operating are accurate and deliberate.
Last updated: 24 September 2026. Effective date: to be set.
Stormbright is operated by Dustorm PreSales, LLC, a limited liability company organised in to be set: state of incorporation. This policy covers the Stormbright web application and nothing else.
Who this policy is for
Stormbright is sold to a company, and used by the people at that company. Those are two different relationships and this policy covers both.
- Your employer is the customer. They create the workspace, decide who joins it, and appoint its administrators. In data-protection language they are the controller of what goes into it.
- You are the user. You upload call transcripts, you get graded, and this policy tells you what happens to the material you put in.
If your employer's administrator can see something, this policy says so plainly. The section Who can read your transcripts is the one most people are actually asking about.
What Stormbright collects
Account information
Your name, your email address, and the workspaces you belong to. Authentication is handled by Google Identity Platform; Stormbright stores an identifier issued by that service and a mirror of your name and email, and never stores your password.
Call transcripts
The core of the product. You paste the text of a sales call into Stormbright and it is stored, graded and kept.
Stormbright stores the transcript as text in its database. It does not record calls, does not connect to your conferencing tool, does not transcribe audio, and has no access to any recording. The only way a transcript enters Stormbright is that a person pastes it in.
Material derived from a transcript
The grade, the coaching artifacts, the demo scripts, and the internal record of each agent run, including the exact prompt sent to the model and the exact response that came back. That internal record contains the full transcript text, because the prompt contains the full transcript text.
Business context
Optional. An administrator may add a description of your company, its products and its differentiators, which the agents read alongside a transcript. It is optional by design and the agents work from the transcript alone when it is absent. See the security page for why.
What Stormbright does not collect
No advertising identifiers, no third-party tracking, no analytics tags, no cookies beyond the session cookie that keeps you signed in, and no sale of anything to anyone.
Where your data is stored
| What | Where |
|---|---|
| Transcripts, grades, artifacts, accounts | Google Cloud SQL for PostgreSQL, region us-central1, United States |
| The application itself | Google Cloud Run, region us-central1, United States |
| Model inference | Google Vertex AI, region us-central1, United States |
| Secrets | Google Secret Manager, replication pinned to us-central1 |
Everything is in one region in the United States. Database backups are pinned to the same region rather than to a multi-region bucket. The database has no public IP address and accepts only encrypted connections.
Encryption at rest is Google-managed AES-256. Stormbright does not currently use customer-managed encryption keys.
If you need data residency outside the United States, Stormbright cannot offer it today. There is one region and it is us-central1.
Who can read your transcripts
This is enforced in the database with row-level security rather than in application code, so it holds regardless of which screen or which query asks.
| Who | What they can read |
|---|---|
| You | Your own calls, transcripts and artifacts |
| Your manager | The same, for the people who report to them, and for the teams they manage |
| A workspace administrator | Every transcript in your workspace, raw and scrubbed, every artifact, and the raw prompts and model responses for every run |
| Another workspace | Nothing. Ever. |
| Dustorm PreSales staff | See below |
Read that third row again, because it is the one that surprises people. An administrator at your own company can read every transcript every colleague has uploaded. That is normal for a business tool bought by an employer, and it is not a defect, but you should know it before you paste a call in.
Dustorm PreSales staff
No part of the access-control system grants Dustorm PreSales staff a route into customer content. There is deliberately no administrative override in the database rules, and an automated test fails the build if one is ever added.
Separately from those rules, Stormbright operates an internal console that runs with a database credential capable of bypassing them. Today that console reads account records, workspace membership and counts, and no code path in it opens a transcript. That restriction is enforced by code review, a lint rule and automated checks rather than by the database itself.
Stated plainly: a person with production database credentials at Dustorm PreSales could read your transcripts. The controls that prevent it are real, but they are procedural and engineering controls, not a cryptographic guarantee. Stormbright does not hold your data in a way that makes it technically impossible for its operator to read.
The scrub, and what it does not do
Stormbright can replace real names in a transcript with fixed labels before the text is sent to a model. It is worth being precise about what this does, because it is easy to over-read.
- It is deterministic text replacement, performed in Stormbright's own process. It is not a model, it does not guess, and using it does not send your transcript anywhere.
- It replaces only the exact terms you list in a substitution table. It does not infer variants, plurals, misspellings or abbreviations. A term you did not enter is not replaced.
- It flags but never replaces email addresses, phone numbers and unrecognised proper nouns, reporting them so you can add them to the table yourself.
- It is optional and off by default. A new workspace has no substitution table. With no table, or with the scrub switched off, or if the scrub is refused because it would have replaced nothing, the transcript is sent to the model as written.
- The original transcript is always stored, unchanged. The scrub produces a second, derived copy. It never overwrites, masks or deletes the original, and the original stays in the database for as long as the call does.
So the scrub governs what the model sees. It does not govern what Stormbright stores. Anyone who can read your transcripts, per the table above, can read the original with the real names in it, along with the table mapping each label back to the name it replaced.
Sub-processors
| Sub-processor | What it does | Sees transcript content |
|---|---|---|
| Google Cloud Platform | Hosting, database, storage, secrets | Yes, as the infrastructure holding it |
| Google Cloud Vertex AI | Model inference | Yes, the text sent for grading |
| Google Identity Platform | Authentication | No |
Models are served through Vertex AI in us-central1. Google's terms for Vertex AI state that it does not use customer prompts or responses to train its models. Stormbright does not restate those terms here, because a paraphrase of somebody else's commitment is worth nothing. The security page links to Google's own published commitments so you can read them at source.
Stormbright does not maintain a separate sub-processor page yet, and there is no mechanism today for notifying customers in advance of a change to this list. to be set: sub-processor change notification commitment
How long data is kept
This section describes an intention that is not yet enforced, and says so rather than implying otherwise.
The intended retention periods are:
| Data | Intended retention |
|---|---|
| Transcripts | 365 days |
| Artifacts and grades | 730 days |
| Raw prompts and model responses | 90 days |
Those figures are a default, they are configurable per workspace, and Duston Littlefield has not confirmed them. Enterprise buyers frequently want 30 or 90 days for transcripts, and changing the default is cheap now and expensive after it has been quoted in an agreement.
No automated deletion is running today. The purge is written and works, but nothing schedules it: the scheduled jobs that would call it are created in a paused state, and the endpoint they would call does not exist yet. Nothing has ever been deleted by a retention process, and no workspace currently has a retention period set in the database at all.
Until that job runs, assume that everything you put into Stormbright is kept indefinitely. This paragraph is the honest answer and it stays here until the job is scheduled and has run.
Deleting your data
There is no self-service deletion today, and no committed response time.
You can delete an individual call from within the product. That marks the call as deleted and removes it from every view. It does not erase the transcript text from the database.
There is no account-deletion flow, no workspace-deletion flow, and no built process for handling a request to erase or export personal data. Such a request is handled manually by an engineer.
Stormbright will not state a response-time commitment it cannot currently meet. When the deletion workflow is built, this section gets a real number. to be set: deletion request response time
To make a request in the meantime, write to to be set: privacy@ address.
Your rights
Depending on where you live, you may have rights to access, correct, delete or export your personal data, and to object to certain processing. Stormbright intends to honour those rights.
As described above, the mechanisms for exercising them are manual today. If you are relying on a specific statutory regime, raise it before you buy rather than after, because the honest answer about tooling is in the section above.
to be set: whether Dustorm PreSales offers a Data Processing Addendum, and on what terms
Security
The technical detail lives on the security page, which is written to be forwarded to an information-security team. In summary: one cloud, one region, one model endpoint, isolation enforced in the database, private networking, encryption in transit and at rest.
Stormbright has not been audited. There is no SOC 2 report, no ISO 27001 certificate, and no penetration test to share. to be set: audit and certification plans
Changes to this policy
When this policy changes materially, the change will be to be set: notification mechanism and notice period. The document's history is tracked in the Stormbright repository, and this page renders directly from that file.
Contact
| Topic | Address |
|---|---|
| Privacy | to be set: privacy@ address |
| Security | to be set: security@ address |
| Legal | to be set: legal@ address |
Dustorm PreSales, LLC, to be set: registered address.