Privacy & data handling
Clear boundaries for business data.
What EkamFlow reads, what it keeps, who can see it, and what you can switch off. This is an overview of how the product behaves, not the agreement itself.
Where your data lives
Read, not uploaded
There is no upload path into EkamFlow. A workflow names tables in the warehouse you connected, and each job — profiling, training, the measurement gate, prediction — reads them through that connection with the credentials you supplied. What it may then keep is the subject of the setting below.
Source-data storage is a switch you hold
By default, EkamFlow may keep source-derived working data: the prepared structure a model trains against, reusable encodings of text it has already read, and, where real-time decisions are set up, the context a decision point needs. Any member can see where this setting stands; only an admin can change it, and every change is recorded.
Switching it off deletes what was stored
Turning the setting off requires an explicit confirmation, then removes the stored working data irreversibly, and from that point refuses the jobs that would write it. Definitions, models, predictions and profiles are not source data and are untouched. Some capabilities — real-time decisions, attribution, factor values in Explainability — need the setting on and say so.
Deleting an organisation
Deleting an organisation removes its records, its scored results and its keys, and from EkamFlow’s own storage the stored copy of its source data, its job settings, its trained models, their checkpoints and the backbones trained for it. The delete then checks that nothing is left, and it is irreversible.
Who can see what
Invite only
Nobody signs up to EkamFlow on their own. EkamFlow creates each customer's workspace and invites the first admin; that admin invites their own team as admins or members. One person can belong to more than one workspace and switch between them.
Access per role, per database, per column
Read access is set per role and per database, and columns can be tagged and hidden from people who should not see them. Masks are applied before a query runs, so a masked column stays masked whatever SQL is written over it.
A key reaches only its own scope
Keys carry a scope — read, predict, profiles, decisions, admin — and are refused on any route outside it. A profiles key reaches a profile's own routes and nothing else; a decisions key reaches one route. Whatever its scope, a key is never treated as an admin when database and column access is decided. Revoking a key stops work already queued under it, not just the next call.
EkamFlow support holds an admin account in your workspace
A support account is added to every workspace when it is created, with the admin role, so the team can help with a workflow that is failing. It can therefore see your data. It cannot be removed, it appears in your member list labelled as EkamFlow's, and everything it does is audited. Ask us about the controls around it during a security review.
Audited
Changes to settings, access, keys and members are recorded, as are the reads an assistant or a key makes.
The assistant and your rows
It acts as the signed-in person
AI Studio's assistant has exactly the access of the person using it, re-checked on every message and every tool call. A database they cannot read does not exist for it, and its name never appears.
Rows reach the model provider
The assistant reads the rows of the queries it runs — that is the point of it — and those rows are sent to the model provider your organisation uses. This is on by default. An admin can switch it off, which leaves the assistant able to discuss the shape of your data but not its values, and can supply your own provider key instead. Ask us which providers are available to you, and how they are covered, before your team starts using it.
Rows stay in their chat
Event records, logs, metrics and error reports hold references, never values. A chat and its dashboards are private to whoever made them. Deleting a chat removes its rows, snapshots and notes at once, and backups of them expire within seven days.
Every outside call is approved
A connector a member adds — their warehouse, their CRM, their own server — asks for approval before each call, and an admin can switch member connectors off or allow only listed hosts.
Whether to trust a model
A model that cannot be measured does not take over
Each use case keeps one champion. A newly trained model replaces it only when it has been measured against the live one on the rows it is entitled to be judged on. A model that does no better than a simple constant is flagged as such.
Runs are bounded
Training and prediction runs carry a cost limit and a time limit. A run that would exceed its limit asks first, in money, rather than spending it.
Scores come with their reasons
Every use case has an Explainability tab showing what drove its scores, read from the model itself — which says what moved the score, not what caused the outcome.
This website
Website enquiries
The demo form collects a name, work email address, company name, and an optional business goal. Alongside it, the enquiry carries how you reached the site: the campaign parameters in the link you arrived through, any advertising click identifier it carried, the page you landed on, and the referring website, where there is one. These details are sent to the EkamFlow team through Resend to handle the enquiry and follow up. The form is for business enquiries: customer records and sensitive account information are not required.
Website measurement
This website uses Vercel Analytics and Speed Insights to understand page usage and website performance. It also records how a visit reached the site: the campaign parameters in the link, any advertising click identifier from Google, Meta, Microsoft, LinkedIn or TikTok that the link carried, the page first opened, and the referring website. The first such visit and the most recent are both kept in this browser's own storage, so an enquiry can be credited to the campaign that prompted it. No cookie is set and nothing is read from other websites, but these are the identifiers an advertising platform uses to recognise a click, and they are kept rather than discarded. They stay in this browser until site data is cleared, and they are sent to EkamFlow when the demo form is submitted; the campaign parameters are also added to the booking link if a time is booked through Calendly, which is a separate service with its own privacy terms.
Policies and agreements
This page is a data handling overview, not the agreement itself. Contact the EkamFlow team for the full privacy policy, service terms, a data processing agreement, or the details needed for a security review.
Contact
For privacy enquiries, the full privacy policy, service terms, a data processing agreement, or the details needed for a security review, contact admin@ekamflow.ai.