This Privacy Policy applies to the Worsley website, browser extension, command-line tool (the “Worsley CLI”), accounts, dashboard, and related services (collectively, “Worsley”). Worsley is currently in beta and offered to early-access customers. By using Worsley, you acknowledge the practices described in this policy.
1. Who we are
Worsley (“Worsley,” “we,” “us”) operates Worsley and is responsible for (the “controller” of) the personal information described in this policy, except where we say otherwise.
For anything about privacy, including requests to exercise your rights, contact [email protected].
2. Information we collect
What Worsley collects depends on which parts of the service you use.
Account and contact information
- Name, email address, account identifiers, and, if you sign in with Google, GitHub, Microsoft, or Hugging Face, the basic profile information that provider shares with us. If you use a password, we store only a secure hash of it.
- A profile picture and project images, if you add them. Worsley stores a small re-encoded copy without embedded information such as where a photo was taken. Members of teams you join can see your profile picture.
- The teams you create or join and your role in each (owner, admin, or member).
- Information you submit through the waitlist, contact, or demo forms, such as your company, role, job title, company size, country, phone number, or use case, when you choose to provide it.
- Communications you send us, including support, privacy, and security requests.
AI sessions captured in your browser (the extension)
- Your prompts and the AI responses shown in the conversation.
- Session and message identifiers, titles, URLs, turns, edits, branches, timestamps, and character counts.
- The AI platform, provider, model, session type or mode (for example a temporary chat), and effort or thinking setting, as the platform reports them.
- Response status and the platform’s own feedback signals, such as a thumbs-up or thumbs-down, when available.
- Attachment details, such as file name, type, size, and source. Worsley never collects the contents of files you attach.
AI sessions captured in your terminal (the Worsley CLI)
The Worsley CLI is an optional program you install on your computer. It captures sessions only from the tools you turn on, and only from the folders you choose when you turn them on. Nothing is read or uploaded unless you’re signed in.
- From terminal and editor AI tools you enable (such as Codex CLI, Gemini CLI, Grok Build, and GitHub Copilot in VS Code): your prompts and the AI’s written replies, session and message identifiers, titles, timestamps, the model, and the tool it came from.
- From AI models you relay with
worsley proxy, whether a model running on your own computer or a provider’s API: your prompts, the model’s replies, earlier messages in the same conversation, and the model’s name. For models on your own computer, it also records token counts and generation times when the model server reports them. - A randomly generated installation ID, the CLI’s version, and your operating system and processor type.
The Worsley CLI does not collect the contents of files a tool reads or edits, commands it runs or their output, tool calls, the model’s hidden reasoning or thinking, system prompts and instruction files, files you reference or attach, your project folders or their paths, your computer’s name, or environment variables. Sessions a tool’s sub-agents run on their own are skipped. Section 3 explains how secrets are removed before anything leaves your computer.
Your choices and feedback in Worsley
- Which chats or AI tools you’ve paused capture for from the extension’s popup. These switches are kept in your browser, not on our servers.
- Your Capture Rules in Settings: whether capture is on, which AI tools are switched off, and whether temporary chats and chats outside a project are kept. These are stored with your account and applied by our servers (section 3).
- Whether you allow Worsley to use your data to improve its own models (off unless you turn it on), with the date and policy version of that choice.
- Your optional answer about whether a session’s task worked out, and any outcome signals you choose to send from tools you connect.
- Which suggestions and insights Worsley showed you, and what you explicitly told us about them (for example “I’ll try this,” “Dismiss,” or a rating).
- Your cookie settings and any opt-out of the sale or sharing of personal information.
Device and technical information
- A randomly generated installation ID for each extension or CLI installation, and the device token that connects it to your account. Each installation has its own token, which you can revoke in Settings.
- Extension and adapter version information, and coarse operating-system and architecture information.
- IP address and request information in server logs, used to operate, secure, rate-limit, and troubleshoot the service.
Information Worsley generates
ROIP scores and the criteria behind them, session types, session metrics such as time spent, prompting insights, trends, and, where model-based features are enabled, estimates such as how a next prompt may score. These are inferences about your prompting, presented to you and, as section 7 explains, in part to your teams.
3. How capture works
Worsley captures in two places: the browser, through the extension, and the terminal, through the Worsley CLI. Both send sessions only to Worsley, over encrypted connections, using a device token that belongs to that installation.
In your browser
- The extension runs only on supported AI services, so it can understand the structure of your sessions. It isn’t built to monitor your general web browsing.
- It captures the reply you see, not the model’s hidden reasoning, and its installation ID isn’t derived from a hardware or browser fingerprint.
- Its toolbar popup shows your scores and suggestions for the current conversation. It doesn’t add anything to the AI service’s page.
- From the top of the popup, you can turn capture off for the current chat or for every chat in an AI tool. While a switch is off, nothing from that chat or tool leaves the page, and turning it back on never captures what you said while it was off.
In your terminal
- Capture is opt-in per tool: a tool you haven’t turned on is never read, and a tool you have turned on is read only from the folders you chose.
- Before anything is saved or uploaded, the CLI removes text that looks like a secret, such as API keys, access tokens, private keys, and passwords in connection strings or configuration lines, and replaces it with a marker. This happens on your computer. Pattern matching can’t recognize every secret, so please don’t paste credentials into AI tools.
worsley proxyaccepts connections only from your own computer, passes every message between your tool and the model unchanged, and records only chat exchanges. A provider API key your tool uses passes through to that provider untouched and is never recorded.- The CLI keeps its sign-in token in your operating system’s keychain, not in a plain file.
You can turn capture off for any tool (worsley disable <tool>), stop the proxy at any time, and sign the CLI out (worsley logout), which revokes its token and deletes everything the CLI kept on your computer. You can disconnect any extension or CLI installation in Settings, remove the extension from your browser, and delete captured sessions at any time.
Capture Rules you set in Settings work differently from the popup’s switches: the extension and the CLI still send what they capture, and our servers apply your rules when it arrives. Anything your rules exclude, such as a chat from an AI tool you’ve switched off there, is discarded without being stored.
Supported platforms and tools may change. See the Supported Platforms page.
4. How we use information
We use information collected through Worsley to:
- Provide, operate, and maintain Worsley, including your account, session history, teams, the extension, and the CLI.
- Grade your sessions, calculate ROIP, and show your prompting patterns, trends, and insights.
- Show team analytics to the teams you belong to (section 7).
- Measure, in aggregate, how often suggestions are shown, accepted, or dismissed and how tasks turn out, to improve them.
- Understand, in aggregate, how Worsley is used, such as sign-ups, active users, account deletions, usage by organization, and website visits, through internal reporting only Worsley staff can access.
- If you opt in, prepare training data for Worsley’s own predictive models (section 6).
- Authenticate users, prevent abuse and fraud, rate-limit requests, and investigate security incidents.
- Debug errors and monitor reliability and performance.
- Respond to support, contact, waitlist, partnership, demo, and privacy requests.
- Comply with legal obligations and enforce our agreements.
We don’t use your information for advertising, and we don’t build profiles to market to you.
5. Legal bases (EEA and UK)
If you’re in the European Economic Area, the United Kingdom, or Switzerland, we rely on these legal bases:
- Performance of a contract: providing your account and teams, capturing and grading sessions you choose to capture, ROIP, insights, and support.
- Legitimate interests: securing Worsley and preventing abuse, keeping it reliable, and understanding and improving it using aggregated measurements. We balance these against your rights, and you can object (section 13).
- Consent: using your data to train Worsley’s own models, and any optional cookies if we ever use them. You can withdraw consent at any time; this doesn’t affect processing before you withdrew.
- Legal obligation: when the law requires us to keep or disclose information.
AI conversations can contain sensitive information. Please don’t capture conversations containing special categories of personal data (such as health information) unless you’re comfortable with Worsley and the service providers in section 8 processing them to provide the service. You can pause capture for any chat or tool (section 3).
6. AI grading, insights, and model training
Your conversations are never used to train generative AI models.
How your sessions are graded
For individual and team accounts, captured sessions are graded by Jev, a decision model from TypeSafe AI, through TypeSafe’s API. Each turn is saved to your account as it arrives, and a session is sent for grading a few minutes after its latest turn, or right away when you close the chat or choose Grade Now, so Jev can read every turn with the rest of the session.
Jev receives the conversation only: your prompts, a shortened excerpt of each AI reply, and the type and name of any attachment (never its contents). It doesn’t receive timing information, your name, or your email address. Jev doesn’t write text, and it doesn’t set your score: it answers a fixed set of questions about each prompt, and Worsley’s own formula turns those answers into your ROIP. Before a session is sent, Worsley checks it for text written to steer the grader; a session that contains such text isn’t sent and is shown as Not Scored.
TypeSafe processes your sessions only to answer those questions, as our service provider, and states that it doesn’t train its models on them. Jev also flags sessions that appear unsafe; flagged sessions are always kept out of any training data.
Insights, when you ask for them
Insights in words, such as what a prompt left out and how to improve it, are generated only when you ask for them. Worsley then sends Claude, a model from Anthropic, through Anthropic’s API, only what that insight needs, without timing information:
- A turn’s insights (for example when you open a turn’s criteria in Session Analysis): your prompt, a shortened excerpt of the reply, attachment details, and what Jev found.
- A session’s insights (Session Analysis › Insights): the session’s title, AI tool, and models; every prompt in the session, shortened; each prompt’s score, what it gave and left out, how each exchange went, and how many files were attached; and your average scores from your other sessions in the last 30 days. Not the AI’s replies.
- Your prompting coach (Actionable Insights › Tips to Improve): figures from your graded sessions in the last 90 days, such as your average ROIP and how it changed, your component averages, ROIP by session type, and how your exchanges went; and, as examples, the lowest-scoring prompt from each of a few of your lowest-scoring sessions and the best prompt from your highest-scoring session, shortened, with those sessions’ titles and scores. Not the AI’s replies.
Claude words the advice; it doesn’t grade. Each result is saved to your account so it isn’t generated again unless you ask for a new one. Anthropic processes this data as our service provider and doesn’t train its models on data sent through its API.
Training Worsley’s own models is opt-in
Worsley may develop small predictive models (for example, to estimate how a next prompt may score or to rank suggestions). They learn only from people who turn on Use My Data to Improve Worsley’s Models in Settings, and only from data captured after they turned it on. Training data is derived from your sessions (scores, measurements, and labels), uses pseudonymous identifiers, and doesn’t include your conversation text.
If you turn it off, delete a session, or delete your account, the affected training data is deleted and kept out of future training. A model that was already trained can’t un-learn it, so those models are flagged for review or retraining. Turning training back on later doesn’t bring back older data. Using Worsley never requires opting in.
Automated evaluation. ROIP scores, insights, and model estimates are automated evaluations of your prompting, provided for your information and, where you’re on a team, your team’s (section 7). Worsley doesn’t use them to make decisions that produce legal or similarly significant effects about you. If an organization uses them in its own decisions about people, that’s the organization’s responsibility under its own policies and the law.
The AI platform you use (for example ChatGPT or Claude) handles your conversation under its own terms and privacy practices; this policy doesn’t change that.
7. Teams and what teammates see
When you join a team, the sessions you capture count toward that team’s analytics, and other people on the team see some information about you:
- Everyone on the team sees your name, profile picture, and role; your number of sessions and time spent; the AI tools and models you use; the names of your projects; and your ROIP, unless the team’s owners have limited scores to owners and admins.
- Team owners and admins can also open each member’s numbers beside the team’s, and see how members compare.
- No one on a team sees what you wrote. Team views show your sessions by number, not by title, and never show your prompts or the AI’s replies.
If you leave a team or are removed from it, you leave its analytics, and your own account and sessions stay as they are.
9. International transfers
Worsley’s servers and service providers are located in the United States. If you’re in the EEA, the UK, or Switzerland, your information is transferred to the United States. We protect those transfers with the European Commission’s Standard Contractual Clauses (and the UK Addendum or Swiss equivalents) or, where a provider is certified, the EU‑U.S. Data Privacy Framework and its UK and Swiss extensions. You can ask us for a copy of the relevant safeguards.
10. Retention and deletion
- Account information: while your account exists.
- Deleting your account closes it right away: you’re signed out everywhere, every extension and CLI connection is revoked, and nothing more is captured. The account and everything it holds is permanently erased 30 days after you delete it.
- Captured sessions and the scores and insights derived from them: until you delete them. A deleted session, including every session removed with Delete All Sessions, moves to Deleted Sessions in Settings for 30 days, where you can restore it or erase it immediately. After 30 days it’s permanently erased.
- Profile picture and project images: until you remove them or delete your account. Replacing one deletes the previous image.
- Training data (only if you opted in): until you turn training off, delete the related session, or delete your account.
- Feedback and suggestion records: while your account exists; deleted with the related session or your account.
- Security log (records of sign-ins, account changes, and other security events, with the IP address and browser details involved): while your account exists; erased when your account is erased. Records not linked to an account, such as a failed sign-in for an email address that has no account, aren’t erased on a schedule.
- Server logs kept by our hosting provider: for a limited period needed for security and troubleshooting.
- Data sent for grading and insights: TypeSafe AI and Anthropic keep it only as long as reasonably necessary to provide their services to us, under their own terms, and don’t train on it (section 6).
- Form submissions and correspondence: as long as needed to handle your request and any follow-up.
Erased information may remain for a limited time in backups before they’re overwritten. We may keep information where the law requires it, and we may keep aggregated or de-identified information that can no longer reasonably identify you.
11. Security
Worsley uses administrative, technical, and organizational safeguards designed to protect information, including encrypted connections, per-user access controls enforced by the database, separate revocable keys for each extension and CLI installation, and restricted access to production systems.
No system can guarantee absolute security. For more detail, see our Security page.
12. Cookies and website analytics
Worsley uses only cookies and browser storage that are strictly necessary for the service to work, and, when you choose a setting like theme or language, preference storage to remember it. The extension also keeps your capture switches in your browser’s storage. We don’t use analytics, advertising, or cross-site tracking cookies. If that ever changes, those cookies will stay off unless you turn them on.
Counting visits to this website. To see which pages people read and how they found us, each page of worsley.ai sends Worsley the page’s address and the page you came from. We keep the page; where the visit came from (a named source such as a search engine or social network, the referring site’s domain, and any campaign tags in the link); your country; whether you’re on a computer, phone, or tablet; and your browser’s name. We don’t use cookies or store anything on your device for this, and we don’t keep your IP address, your browser’s full details, or the full address of the page you came from. Visits on the same day are told apart by a code made from your IP address, your browser, and a random value that’s deleted after two days, so a visit can’t be linked to you or to your visits on other days. Visits to your account and the Worsley app aren’t counted this way.
| Name | Category | Purpose | Provider | Duration |
|---|---|---|---|---|
| __Host-worsley-session_token | Essential | Keeps you signed in. HTTP-only; not readable by scripts. | Worsley | Until you sign out, or after 7 days without use |
| __Host-worsley-dont_remember | Essential | Ends your session when the browser closes if you chose not to stay signed in. | Worsley | Browser session |
| __Host-worsley-signup_binding | Essential | Protects sign-up and sign-in from cross-site request forgery. | Worsley | Minutes |
| worsley_privacy | Essential | Remembers the choices you make here. | Worsley | 12 months |
| worsley.signedIn | Essential | Tells the app you're signed in so it opens the right screen. Contains no personal information. | Worsley | Until you sign out |
| worsley.reloadedForUpdate | Essential | Set when a page reloads to pick up a new version of Worsley, so it never reloads in a loop. Contains only a time. | Worsley | Until you close the tab |
| Demo request progress | Essential | Keeps what you typed in the demo-request form while you move between its steps. | Worsley | Until you close the tab or submit |
| Cloudflare security cookies (e.g. __cf_bm, cf_clearance) | Essential | Protect Worsley from bots and attacks; set only when Cloudflare's security checks require them. | Cloudflare | Up to 30 minutes (cf_clearance up to the configured challenge period) |
| Cloudflare Turnstile | Essential | Human verification on sign-up and sign-in. Not used for advertising. | Cloudflare | Browser session |
| worsley-theme | Preferences | Your light, dark, or automatic theme, if you pick one. | Worsley | Until you clear it or switch preferences off |
| worsley-locale | Preferences | Your language, if you pick one. | Worsley | Until you clear it or switch preferences off |
| worsley-motion | Preferences | Your reduced-motion setting in the app, if you pick one. | Worsley | Until you clear it or switch preferences off |
You can review or change your choices at any time:
Global Privacy Control and Do Not Track. If your browser sends a Global Privacy Control signal, we treat it as a request to opt out of the sale or sharing of your personal information. If it sends either Global Privacy Control or Do Not Track, worsley.ai doesn’t count your visits at all. We don’t track you across websites in any case.
13. Your rights and choices
Depending on where you live, you may have the right to:
- Access your personal information and get a copy in a portable format.
- Correct inaccurate information.
- Delete your information.
- Object to, or ask us to restrict, certain processing, including processing based on legitimate interests.
- Withdraw consent at any time, such as turning off model training in Settings.
- Opt out of the sale or sharing of personal information (Worsley doesn’t sell or share it).
Many of these you can do yourself in Worsley: pause capture from the extension’s popup or the CLI, export your sessions, delete or restore sessions, change training settings, disconnect devices, leave a team, or delete your account in Settings. Otherwise, email [email protected]. We’ll verify your identity before acting (usually by confirming control of your account email) and respond within one month in the EEA and UK, or 45 days under US state laws, extendable where the law allows. You can use an authorized agent, who must show your written permission. We won’t discriminate against you for exercising your rights.
Appeals and complaints. If we decline your request, you can appeal by replying to our decision; we’ll respond within the time the law requires. If you’re in the EEA or UK, you can also complain to your data protection authority (in the UK, the Information Commissioner’s Office). We’d appreciate the chance to address your concern first.
14. US state privacy notice
This section supplements the policy for residents of California and other US states with comprehensive privacy laws (including Colorado, Connecticut, Virginia, Utah, Texas, Oregon, and others).
Categories we collect, and why
- Identifiers (name, email, account and device IDs, IP address): to provide and secure the service.
- Customer and professional information (company, role, job title from forms; team memberships and roles): to respond to requests and provide team features.
- Internet or electronic network activity (AI sessions captured in the browser and the terminal, usage of Worsley, feedback): to provide grading, ROIP, insights, and analytics.
- Inferences (ROIP scores, session types, insights, estimates): presented to you about your prompting, and in part to your teams.
- Sensitive personal information: account login credentials, and the content of conversations you choose to capture. We use it only to provide the service you request and as otherwise permitted by law, not to infer characteristics about you, so the right to limit its use doesn’t apply.
We collect this information from you, from your browser, the extension, and the CLI, and from sign-in providers you choose. We disclose it for business purposes only to the service providers listed in section 8 and to your teams as described in section 7. In the past 12 months, we have not sold or shared personal information, and we don’t knowingly sell or share the information of anyone under 16. Retention is described in section 10.
Your rights include knowing, accessing, correcting, and deleting your information, opting out of sale, sharing, or targeted advertising, and not being discriminated against. To exercise them, see section 13. You can record an opt-out of sale or sharing here:
15. Third-party AI platforms
Worsley works alongside third-party AI products and tools, such as ChatGPT, Claude, Gemini, and Copilot in the browser, and terminal tools and model APIs in the terminal. Those services are operated by their providers, not by Worsley, and their handling of information is governed by their own terms and privacy policies. The same is true of the provider whose API key you use through worsley proxy.
Worsley’s support for a platform doesn’t mean the platform provider sponsors, endorses, or is affiliated with Worsley unless we expressly say so.
16. Age requirement
Worsley is for adults. You must be 18 or older to create an account or use Worsley, and it isn’t directed to anyone under 18. We don’t knowingly collect personal information from anyone under 18. If we learn that we have, we’ll delete it and close the account. If you believe someone under 18 has provided us information, contact [email protected].
17. Changes to this policy
We may update this policy as Worsley changes, and we’ll update the effective date at the top. If a change materially affects how we handle your personal information, we’ll give you notice in advance, for example by email or in the app, and we’ll ask for your consent again where the law requires it.
18. Contact us
For privacy questions or requests, contact:
For anything else, use [email protected] or the Contact page.