Privacy Policy
What the Workforce Intelligence connection service, Workforce Intelligence inference service and account portal, and documentation site do with personal data, and the choices you have.
Who we are#
This service is operated by Workforce Intelligence, Inc. ("we", "us"). Our postal address is 1 Seneca St, Floor 24, Buffalo, NY 14203. You can reach us at privacy@getwi.com.
What this service is#
This policy covers three things we run:
- The connection service. A customer's own Workforce Intelligence installation uses it to connect to accounts at the outside services that your organization's installation connects through this service, without registering its own application with each one. When a person chooses to connect an account, the connection service completes the sign-in step with the provider of that service and hands the result back to the installation that asked for it.
- The Workforce Intelligence inference service and account portal. A customer's own Workforce Intelligence installation can send requests for model responses to the inference service (the programming interface at
/v1), signed with an API key from the customer's Workforce Intelligence account. The account portal (the website at/portal) is where a person creates and manages that account: sign-in, card, API keys, monthly spend limit, usage statements and invoices. - Optional bug reports from an installation. An installation whose administrator has turned on bug-report sharing can send us a short, fixed-format report about a fault, as described below. By default a report holds only fixed fields; an administrator may also choose to include text written by a person, a Coworker or an app. Nothing is sent unless that choice has been made.
- Deployment identity for installations. An installation proves who it is to us with a key pair, as described below, so that we can tell installations apart and stop one that misbehaves.
- The documentation site. The public user and administrator guides for Workforce Intelligence.
This is not the customer's Workforce Intelligence installation. Each installation has its own operator, and that operator's own privacy policy covers the data the installation holds, including the content of connected accounts. The connection service never receives that content. The inference service receives only the content of the requests an installation sends to it, as described below.
What data passes through the connection service#
When a person connects an account, or an installation refreshes an existing connection, the following passes through the service:
- The authorization code the provider issues after the person approves access.
- The access token and refresh token the provider issues in exchange for that code, and the new tokens issued when a connection is refreshed.
- The web address of the installation that started the connection, so we can send the person back to it.
- The provider's name and the permissions (scopes) requested.
- A connection-attempt identifier and a one-way fingerprint created by the installation, which let the installation prove the result is meant for it.
We use this data only to complete the connection or refresh that the person's installation asked for.
What we do not do with connection data#
This section is about the connection service.
- We do not keep access tokens or refresh tokens. The only place a token is held, even briefly, is a sealed, encrypted hand-off record. It is accepted for at most 5 minutes and is deleted the first time the installation collects it. A refresh stores nothing at all.
- We do not read the content of connected accounts. We never call a provider's data interfaces such as mail, files, calendars or messages. The only provider calls we make are the sign-in token exchange and the token refresh.
- We do not sell or share this data.
- We do not use it for advertising.
- We do not use it to create, train or improve any machine learning or artificial intelligence model.
The Workforce Intelligence inference service: requests and responses#
When an installation sends a request to the inference service, we receive its content: the messages, system instructions, tool definitions, and any images or files it contains. We use that content for one purpose, to produce the response.
- We do not store prompts or responses. Request content and the response are held in memory only while the request runs. No prompt or response is written to our database, our accounting exports or our logs. The record we keep for each request holds usage details only, listed in the next section.
- We forward the request to a model provider. The request content is sent to the model provider chosen to answer it, and the response is streamed back. When the provider of the requested model cannot answer (it is unavailable or overloaded, it rejects our credentials, or our account with it cannot be used), we can send the same request to another provider, but only before any part of a response has been sent. For a model family name, that means the same model through Amazon Bedrock where Bedrock offers it, or an equivalent model of the same level (low, mid or high) from a different provider, so the model that answers can come from a different provider than the name suggests. Your usage statement shows how many requests were answered this way, and no request is sent to a provider other than the ones named below. A request can therefore reach more than one provider. The providers we send requests to are named under "Who receives data".
- We send only what the request contains. We add no name, email address, account identifier or API key of the customer to what a provider receives. The provider receives the request content, the output limit and our own credentials for that provider.
- We do not use request content for any other purpose. We do not read it, sell it, share it except with the providers described above, use it for advertising, or use it to create, train or improve any model.
- What the model providers do is theirs to state. Each provider handles what it receives under its own terms and privacy policy, including how long it keeps it and what it uses it for. We do not control that.
The inference service does not connect to any outside account. It receives only what the installation sends.
The Workforce Intelligence account portal: account data#
To run a Workforce Intelligence account we hold the following.
- Sign-in details: the email address, and the password as a one-way argon2id hash. We never store the password itself.
- Sessions: a record of each signed-in browser, found by a one-way hash of its cookie, with its times and a security token. A session ends after an hour without use and after 12 hours in any case.
- Email confirmation and password reset links: only a one-way hash of the current link and its expiry. A confirmation link works once and for 24 hours; a reset link works once and for one hour.
- Billing state: the identifier Stripe gives the customer, whether a card is on file, and one invoice record for each month: amounts, status and the Stripe invoice identifier. Card numbers are entered on a page hosted by Stripe and never reach us. The card's brand and last four digits are read from Stripe when we show them and are not stored.
- API keys: the first part of each key (its prefix), a one-way argon2id hash of the rest, the name you gave it, when it was created and last used, and whether it is revoked. The key itself is shown once, when you create it, and we cannot show it again.
- The monthly spend limit you set, and a record of changes to the account (who changed what, which setting and when, by identifier). These records hold no email address, password or key.
- Usage records: for each request, its identifier, the model name you asked for, the model that answered and its provider, token counts (cached tokens included), the amounts billed and the pricing details used to work them out, time, status, how long it took and the prefix of the key used. Month totals are kept per account. No prompt, response or file is in these records.
- Operator records: when one of our staff with operator access views, exports or changes accounts or billing, we record who, what, which account and the outcome, by identifier. They never include the operator's secret code.
- Email we send: a confirmation link, a password reset link, and notices about the account, such as a password change, a suspension or its end. They are sent through Amazon Simple Email Service. We use email only for these purposes, never for marketing, and we do not keep a copy of the messages.
- Cookies: the portal sets a session cookie and a security (CSRF) cookie. Both are needed for sign-in and for the forms to work, are marked secure and not readable by scripts, and are not used for tracking. The portal has no analytics or advertising.
Optional bug reports from an installation#
A Workforce Intelligence installation can send us a bug report, but only when an administrator of that installation has turned bug-report sharing on. It is off by default, an administrator can turn it off again at any time, and an installation that has it off sends nothing. We also have to switch on receiving at our end; until we do, the address that accepts reports answers as if it did not exist.
What a report contains is a fixed list of fields. By default there is no free text, so no typed words, and nothing from a conversation, document or file, is in a report:
- A report identifier (a random value that identifies the report so that one report is not stored twice) and the time the report was filed.
- The kind of sender: a person, a coworker, an app or a service. This is a category only, never a name, email address or account identifier.
- The category of the problem, chosen from a fixed list (for example an empty model response, a tool error, or trouble finding something).
- The Workforce Intelligence version the installation is running.
- For an empty model response only: the provider name, the name of the model that was asked for and of the one that answered, why the response stopped, the token counts the provider reported, and the identifier of the technical trace of that request (a random value that is not a prompt or a response).
Apart from the optional text described below, a report never contains a prompt, an instruction, a response, a file or a document, a name, an email address, a page address, the name of a connected account, a credential or a key, or the address of the installation. We check every report against the fixed list when it arrives, and refuse any report that holds something else rather than keep it.
Optional text. By default a report holds fixed fields only. When a person, a Coworker or an app files a report, an administrator of the installation may choose to include text that it wrote for that one report: a summary (a description of at most 2,000 characters) and, for a report from a Coworker or an app, a detail (plain text of at most 4,000 characters). The installation shows the administrator exactly what would be sent, and the administrator reviews it, and can edit it, before it is sent; nothing is sent without that approval. Reports from automatic services never carry this text. Because it is whatever the person, Coworker or app wrote, it can contain anything, including personal data or business information, so the administrator should remove anything they would not want us to read. Our staff with operator access can read it. It is kept for 180 days with the rest of the report and then removed by the database (see "Retention"). To ask us to delete it sooner, contact us as described under "Contact" and give the report identifier shown in the installation.
We use reports to find and fix faults in Workforce Intelligence. Only our staff with operator access can read them. We do not share them with anyone else or use them for any other purpose. They are kept for 180 days and then removed by the database (see "Retention").
Like any request to our service, a report arrives with the sending server's IP address, which is covered under "Network metadata". We do not write the contents of a report, including this text, to our logs.
Deployment identity#
A Workforce Intelligence installation enrolls with us automatically the first time it starts. It creates a key pair, keeps the private key to itself, and sends us the public key with a signature that proves it holds the private key. Together with its identifier, the installation also sends us the company name set up in it, so that our staff can tell whose installation a request came from. We store, for each installation, the public key, an identifier derived from it, the company name the installation sends (shown only to Workforce Intelligence operators and never used to decide what an installation may do), the time it enrolled, the time we last heard from it (updated at most once an hour) and whether an operator has revoked it. A public key is not a secret and is not personal data about anyone. The company name is information about the organization, not about a person. We do not receive or store the installation's host name, address, users or any other customer data for this purpose.
The installation then signs its calls to our bug-report endpoint and to the connection service's token refresh, so that we know which installation sent a call and can refuse one that is unsigned, stale, replayed or wrongly signed. A token refresh is not tied to an installation; only its signature is checked. An operator can revoke an installation, after which it is refused at once; our staff with operator access can see the list of installations and their status. We use this only to run, secure and fix the service. The data is kept until an operator removes it (see "Retention" and "Backups and exports").
Operational logs#
We keep operational logs to run and protect the service. A log line from the connection service contains the request path, request method, response status, how long the request took, the provider's name, the outcome of the step and a stable error code or error type. Logs never contain tokens, authorization codes, client secrets, handoff identifiers or query strings. The connection routes are also excluded from request tracing.
Log lines from the inference service and the portal contain the same kind of facts: the path, method, status and duration of a request, the request's identifier, account identifier, model name, provider, token counts, amounts, status and error code or type. They never contain prompts, responses, API keys, passwords, confirmation or reset links, or query strings, and the confirmation pages are never traced.
Logs are kept for one month and then deleted.
Network metadata#
Our cloud hosting provider and its web firewall process the IP address and request details of every request in order to deliver the service and block abuse. We have not turned on load balancer access logs or web firewall request logging. The web firewall retains a small sample of recent requests, including source address and headers, for three hours, and records request counts as aggregate metrics.
The documentation site#
The documentation site is static. It sets no cookies (the cookies above belong to the account portal), includes no analytics or advertising, and loads nothing from third parties. If you choose a light or dark theme, that choice is stored in your own browser's local storage and is never sent to us. Search runs in your browser against an index served by the site.
Why we process data, and our legal bases#
- To provide the connection service a customer's installation requested. The bases are performance of our agreement with the customer and the legitimate interest of the person connecting their own account.
- To run a Workforce Intelligence account, answer requests sent to the inference service, bill for them and show statements. The basis is performance of our agreement with the customer.
- To keep the service secure, prevent abuse and fix faults. This includes request-rate limits, spend limits, audit records of account and operator actions, and the bug reports an installation's administrator chooses to send, with written text only when the administrator approves it. The basis is our legitimate interest in running a secure and reliable service.
- To keep accounting records and meet legal obligations, where the law requires it.
We do not use consent as a basis, because we do not collect data for purposes that need it.
Retention#
| Data | How long it exists |
|---|---|
| Pending connection attempt record (no personal data, an unguessable identifier only) | Rejected after 10 minutes; removed when the provider returns the person |
| Sealed hand-off record holding the result | Rejected after 5 minutes; deleted on the first collection attempt |
| Tokens during a refresh | Held in memory for the request only; never stored |
| Operational logs | One month |
| Bug reports sent by an installation that has turned sharing on | 180 days after we receive them, then removed by the database |
| Deployment identity of an installation (public key, derived identifier, company name, enrollment and last-seen times, status) | Kept until an operator removes it. Nothing deletes it automatically |
| Web firewall request samples | Three hours |
| Prompts and responses sent to the inference service | Held in memory while the request runs; never stored |
| Usage records, month totals and invoice records | Kept as accounting records. Nothing deletes them automatically |
| Email address and password hash of an account | Kept while the account exists; removed when we act on a deletion request (see "Your rights") |
| Sessions | Rejected after an hour without use and after 12 hours; the database removes them later |
| Confirmation and reset links (hash only) | Usable once, within 24 hours (confirmation) or one hour (reset) |
| API key records (prefix and hash, no secret) | Kept after a key is revoked, marked revoked |
| Account change and operator records | Kept. Nothing deletes them automatically |
| Payment event records (event type and state only) | 90 days, then removed by the database |
| Daily accounting export of the account and usage table | Not deleted automatically (see "Backups and exports") |
Expired records are rejected as soon as they expire. Our database then removes them physically; Amazon Web Services documents that it deletes expired items "typically within a few days after their expiration". Until a record is removed it remains encrypted.
Backups and exports#
The connection service's hand-off table has no backups. The bug-report table has no backups either, so a report is gone from our systems once it expires.
The table that holds deployment identities has point-in-time recovery, which lets Amazon Web Services restore it to any second in the previous 35 days. It holds public keys, identifiers and company names only. We keep it so that losing it would not make every installation enroll again or make us forget which ones were revoked.
The table that holds Workforce Intelligence accounts, usage records and invoice records has point-in-time recovery, which lets Amazon Web Services restore it to any second in the previous 35 days. A record changed or removed in that table can therefore still be restored from that history for up to 35 days.
Once a day we also export the whole table to a private, encrypted storage bucket for accounting and reporting. Each export holds the same records as the table on that day, including account email addresses and password hashes, and never any prompt or response. The bucket keeps previous versions of replaced or deleted files for 90 days, moves exports to colder storage after 90 days, and has no rule that deletes an export.
Who receives data#
- Our hosting provider, Amazon Web Services, which hosts the service, its databases, its storage, its secrets store, its web firewall and the email service (Amazon Simple Email Service) that delivers our messages.
- The outside service the person chooses, whose provider receives the sign-in exchange and refresh requests necessary to issue tokens.
- The installation that started the connection, which receives the tokens in the final step.
- Stripe, which collects the card, keeps the customer record (we give it the account's email address and our account identifier), issues invoices and charges the card. Stripe handles that data under its own terms and privacy policy.
- The model providers that answer requests: Anthropic, OpenAI, Google (for its Gemini models), xAI and Amazon Web Services through Amazon Bedrock, which runs models made by Anthropic, OpenAI, xAI, Cohere and Stability AI. They receive the content of the requests sent to them, as described above, and handle it under their own terms and privacy policies.
- Our staff with operator access, who can see accounts (including the email address), usage and billing to run the service and bill customers, who can read the bug reports an installation has chosen to send, and who can see the installations that enrolled and revoke one.
We do not sell personal data or share it with anyone else, except where the law requires it.
International transfers#
The connection service and the account and usage data we hold are hosted on Amazon Web Services in the United States, and that data is stored and processed only there. Requests we send to Amazon Bedrock are processed by Amazon Web Services in the United States: we use Bedrock only in Amazon Web Services regions in the United States and through its inference profiles for the United States, which keep requests within the United States. The other model providers and Stripe process the data we send them in their own locations, which can be outside the United States. If you are outside the United States, your information is transferred to and processed in the United States when you use the service.
Security#
The hand-off records are encrypted, connections use HTTPS only, and requests are validated. Passwords, API keys, session cookies and confirmation and reset links are stored only as one-way hashes. The tables are encrypted at rest and the export bucket is private and encrypted. The rules for using the service are in the Terms of Service. Data handling and security describes each measure.
Your rights#
Depending on where you live, you can ask us to give you access to your personal data, correct it, delete it, restrict or object to our processing of it, or provide it in a portable form. You can also complain to your local data protection authority.
Because we do not store tokens beyond the short hand-off window, a request about the connection service is answered by confirming that we hold no personal data about you beyond operational logs and network metadata kept for the periods above.
If you have a Workforce Intelligence account, we hold the account data listed above. In the portal you can see and download your usage statements, create and revoke API keys, change the spend limit, reset your password and sign out everywhere. The portal has no self-service way to delete an account; to correct or delete your account data, or to make any other request, write to privacy@getwi.com. We handle these requests by hand. When we delete an account's data, we remove the sign-in details (email address, password hash and sessions) and replace the email address on the account record, and we keep what accounting requires: usage records, invoice records and the records of account and operator actions, which identify people only by opaque identifier. Copies in the daily accounting export are described under "Backups and exports". Stripe holds the card and customer record under its own deletion process.
To remove an installation's access to your account, see the section on revoking access in Data handling and security. To stop an API key working, revoke it in the portal; it stops working within a minute.
Google user data#
If you connect a Google account through this service, the following applies to information received from Google.
The use of information received from Google Workspace scopes will adhere to the Google User Data Policy, including the Limited Use requirements.
The use and transfer by Workforce Intelligence, Inc. to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements, and to the Google Workspace API User Data and Developer Policy.
What Google data we access. We access only what Google returns when the person approves the sign-in: an authorization code, and the access token and refresh token issued for it. The permissions (scopes) requested are chosen by the customer's installation when it starts the connection, and the service accepts only scopes from the list it supports for that provider; the person sees them on Google's consent screen before approving. We do not use those tokens to call Google APIs. We never read the content of the account, such as mail, files or calendars, and we take no actions on the person's behalf.
How we use it. We use Google user data only to provide the connection a customer's installation asked for: completing the sign-in, delivering the resulting tokens to that installation, and refreshing those tokens when the installation asks. This is a user-facing feature of the installation the person chose to connect.
How we store it. We do not store Google user data beyond the sealed, encrypted hand-off record described above, which is accepted for at most 5 minutes and deleted the first time the installation collects it. A refresh stores nothing. Tokens are encrypted in transit over HTTPS and at rest while held in the hand-off record.
How we share it. We do not transfer Google user data to others, except to the installation that started the connection, as necessary for security or abuse investigation, or to comply with law. We do not sell it, and we do not transfer it to advertising platforms, data brokers or information resellers.
Humans and models. We do not allow humans to read Google user data. Tokens are sealed and we do not open them for review. We do not use Google user data for advertising, and we do not use it to create, train or improve any machine learning or artificial intelligence model.
Retention and deletion. Google user data exists in the connection service only for the periods in the retention table above. Because nothing is kept beyond them, there is nothing further to delete; to remove an installation's access to your Google account, revoke it in your Google Account permissions, as described in the section on revoking access in Data handling and security. You can also write to privacy@getwi.com with any request.
Data from other outside services#
For any other outside service an installation connects through this service, we use data from the person's account only to provide the connection the customer asked for, under the same limits as for Google: it passes through the service during connecting and refreshing, we do not read the content of the account, we do not store tokens beyond the sealed hand-off record, and we do not sell, share, advertise with or train models on it.
Children#
The service is not directed to children under 16, and we do not knowingly collect personal data from them.
Changes to this policy#
When we change this policy we update the "Last updated" date above. If a change affects how we use data the service already handles, we publish it before it takes effect and change the "Effective" date.
Contact#
Workforce Intelligence, Inc.
1 Seneca St, Floor 24, Buffalo, NY 14203
privacy@getwi.com