Data Use Policy
AgenticED, Inc.
Effective Date: October 13, 2026
Supersedes: Data Use Policy dated January 6, 2026
Why This Document Exists
The Privacy Policy covers what we collect, who processes it, and how long we keep it. This document answers a narrower question that an AI platform owes its districts a direct answer to:
When the platform gets better, where does the improvement come from, and who owns it?
For many AI vendors, customer data improves a model everyone uses. That is how the vendor's product improves, and it is how one district's data ends up working for the next district. Our answer is different, and it comes from how the platform is built.
The Rule
What the platform learns from your district stays with your district.
A district's data is held inside the district's own tenant boundary, with row-level security enforced in the database. A pattern learned from your students is a trait of your district's configuration. There is no mechanism by which it becomes a platform-wide capability, and there is no mechanism by which it reaches another district.
Three commitments follow:
We do not train AI models on your data. Not our own models and not our providers' models. We use our language-model providers under commercial terms that do not permit them to train on the content we send.
We do not pool district data. We do not combine data across districts to produce a benchmark, a model, a norm, or a product, and we do not sell or publish aggregate findings derived from your students.
We do not use your data to sell you more. A district's data does not feed our product roadmap, our marketing, or our sales.
We improve the platform continuously. What we learn from is how the platform itself performs: which agent requests fail and why, where a workflow stalls, which screens people leave, where a translation is missing, and how long an eligibility estimate takes. That is operational data about our software. It is not analysis of your students.
How We Use Your Data
Only as your district authorized. In practice:
Operating the service. Running the agents your district enabled, at the scope it set, and serving the screens your staff and families use.
Estimating what a family might be eligible for. Benefits and aid eligibility, calculated inside our own infrastructure. See Where the Calculation Happens, below.
Showing signals to the people entitled to see them. A counselor's caseload, a teacher's roster, a principal's early-warning tile, each limited by role, consent, and the district's configuration.
Keeping the record. The audit, consent, and disclosure records your district relies on for its own compliance, so your district can show who changed or received what.
Supporting you. Diagnosing a problem you report, which sometimes requires looking at the data involved. We do it at your request, and we tell you.
Uses We Never Make of It
- Training or fine-tuning any AI model
- Advertising, targeted advertising, or advertising profiles
- Sale, licensing, or transfer to a data broker
- Cross-district benchmarking, norming, or research without the district's specific written authorization
- Any secondary purpose the district did not authorize
- Our own sales, marketing, or product development
Where the Calculation Happens
Benefits eligibility is estimated by a rules engine we run as our own service inside our own private network. Household composition and income ranges are evaluated there.
We do not call the rules engine's public service. If our own instance is unavailable, the platform declines to calculate instead of sending household information outside our network. A family may have to wait, and their household information stays inside our network.
Local rules, such as a city, district, community organization, or utility program, are held as data in our own tables. A local eligibility rule never becomes a modified copy of a national one.
How the Knowledge Layer Works
Our agents answer from a knowledge library, because a model's general memory of a benefit threshold is a guess delivered with confidence.
Verified amounts come from verified sources. Federal aid maximums, loan limits, income thresholds, and program amounts are served from stored values taken from the agencies that publish them, and we update them when those agencies change them. That is why our agents can state a number.
District content answers district questions. Where a district provides its own policies, deadlines, and program information, our agents answer from that content.
Academic standards are public reference data. Standards structure is identical for every district, and a request for it carries only a public standard code. No student or district information leaves our network with it. If the outside source is unavailable, we serve our stored copy and label which source answered.
When we do not know, we say so. An eligibility question we cannot answer is reported as undetermined, never as a "no." For a family, "we could not tell whether you qualify" and "you do not qualify" lead to opposite actions, and we will not let a gap in our data produce the wrong one.
District-Controlled Infrastructure
A district can supply its own infrastructure for the parts of the platform where its control matters most:
| Option | What it means |
|---|---|
| Your own key | Per-district key management, designed so the platform cannot decrypt a district's data on its own, with a process for moving existing data under your key. Available on request. |
| Your own email identity | A district may send platform email under its own sending identity, so messages to its families come from the district. A district that does not set one up sends under ours. |
| Your own systems of record | Your student information system, your Ed-Fi instance, and your rostering provider remain yours. We read from them. We do not replace them. |
A district that cannot leave was never in control. Being able to take your data and go is part of the product.
Pseudonymized and Longitudinal Data
JustData produces pseudonymized and longitudinal views for a district's own analysis.
We treat that data as fully protected. Our implementation keeps a re-identification key, a way back from a token to a student, because analysis across years requires the link to persist. Because we hold the key, the data is pseudonymized, and under a conservative reading of FERPA pseudonymized data remains an education record. We apply the same protections to it, we log every use of the key, and re-identifying a record requires a separately authorized and logged action.
A district's pseudonymized data is still the district's data. It does not enter a cross-district pool, and it does not become a benchmark we sell.
Workforce Data
Where a district enables our workforce products, employee signals have their own rules, enforced in the database: administrators see aggregates only; an aggregate is suppressed entirely below a minimum group size the district cannot lower; and a request made for evaluation purposes is refused with an error.
Employee data is never used for evaluation, discipline, or an employment decision, by us or through our platform. A district's collective bargaining agreement applies alongside this policy, and where it is stricter, it controls.
We administer no employee survey. We do not collect engagement or burnout survey responses, and no survey response store exists in the platform.
See the Security Overview for how each of these rules is enforced.
Research
We do not conduct research on district data, and we do not permit a third party to, without the district's specific written authorization for a specific purpose. Accepting our Terms is not that authorization. Where a district authorizes research, it may review any findings that describe its students before they are published.
Changes
We will not expand our use of your data beyond what this policy describes without your district's consent. We will post changes with a new effective date and notify districts directly at least thirty days before a material change takes effect.
Contact
AgenticED, Inc.
Attn: Chief Compliance Officer
240 West 136th Street
New York, NY 10030
privacy@agenticed.com