Privacy Policy
SpecHawk Inc. ("SpecHawk," "we," "us") makes SpecHawk, a QA platform that connects to web applications you choose, maps how they work, writes and runs tests, and reports the results with evidence. This policy explains what we collect, how we use it, who we share it with, and the choices you have. It applies to spechawk.ai and the SpecHawk service.
If your company has a signed agreement with us, such as a design partner agreement or an order form, that agreement controls how we handle your company's data where it differs from this policy.
1. The two kinds of data we handle
Account and website data is information about you and how you use our website and service. We decide how it is used, and this policy describes that.
Customer Data is what your company connects to SpecHawk, and what SpecHawk captures from the environments you ask it to test: environment addresses, sign-in details, page content, screenshots, recordings, logs, test steps, results and findings. We process Customer Data on your company's behalf, to provide the service to your company.
2. What we collect
Account details. Your name, work email, company and role in your workspace, and your password, which we store only as a one-way hash. If you sign in with Google, we get your name and email address from Google instead, and no password.
Environment setup. The web addresses you add and any sign-in details you give us, such as usernames and passwords, saved session cookies, extra login fields and proxy settings. We encrypt sign-in secrets before we store them and decrypt them only when SpecHawk signs in to an environment: to map it, run a test or check that sign-in works. SpecHawk types a saved password into your sign-in page itself. It is never shown to an AI model, and the product never shows a saved password back to you.
What SpecHawk sees while it works. When SpecHawk maps an environment or runs a test, it opens pages in a browser the way a person would and records what it needs to show you results: page addresses, page text and structure, screenshots, video recordings of runs, console messages, failed requests, the cookies and headers a site sends, and the code libraries a page loads.
Emailed sign-in codes. An environment that signs in with emailed one-time codes gets its own inbox address on our domain. For each message sent to that address we keep the sender, the subject and the time it arrived, and we store the message body and the code encrypted. We delete each message 24 hours after it arrives. Messages that fail spam or virus checks are dropped and not stored.
Connected tools. If you connect Jira, we read the tickets you choose to turn into tests. If you link a GitHub repository, we copy it for the length of a scan and read its routes, file paths, frameworks and the names and versions of its packages. We delete the copy when the scan ends and keep only what we read. We do not store your source code.
Usage and device data. Your IP address and browser type, which reach our servers and our network provider with every request, and error logs. We do not run website analytics or tracking scripts.
What you send us. Messages you send us, and your answers when you apply to our design partner program.
3. How we use it
- To provide the service: sign you in, map environments, write and run tests, store results, run schedules and send the notifications you ask for.
- To keep SpecHawk and your data safe: detect abuse, investigate problems and prevent fraud.
- To support you, and to improve SpecHawk using de-identified, aggregated usage information that does not identify you or your company.
- To improve SpecHawk's AI agents, using the session records described in section 4.
- To contact you about your account and important changes, and, if you agree, about news. You can opt out of marketing emails at any time.
- To meet legal obligations.
We do not sell personal information, and we do not share it for targeted advertising.
4. AI features
SpecHawk uses large language models to map your app, write tests from plain English or tickets, name and group the workflows it finds, repair test steps that stopped working, review screenshots and conversations for visual and conversation checks, and answer questions in the Hawk, our assistant. To do this, we send the relevant content, such as page structure, page text, a screenshot and your instructions, to an AI model provider, which processes it and returns a result.
On SpecHawk's own key, requests go to Anthropic's Claude models, through the Claude API and Claude Managed Agents. Until a workspace picks one, Agent UI checks use Anthropic's Claude Sonnet 5.5 and everything else uses Claude Haiku 5.5, both through the Claude API. If your workspace picks a model that isn't Claude from our list (DeepSeek, Z.ai, OpenAI, Qwen, Meta's Llama, Google's Gemma or Mistral), its calls other than Agent UI checks go through OpenRouter to that model. We ask OpenRouter to send requests only to providers that don't keep the content and don't train on it. By default, Anthropic doesn't use anything we send through its API to train its models. It deletes API inputs and outputs within 30 days by default, and it may keep content its safety systems flag for up to 2 years. Longer retention applies to services we control: Anthropic keeps Claude Managed Agents sessions until we delete them. If your workspace adds its own key for OpenRouter, Anthropic or OpenAI, requests go to that provider instead.
We do not use Customer Data to train AI models, ours or anyone else's. What a model provider may do with the content it receives is set by its own terms. If your workspace adds its own AI provider key, requests made with that key are covered by your own agreement with that provider.
Session records. We keep a record of each AI session: what we asked the model, what it answered, and what the agent saw and did in your test environment, such as the pages it opened, the screenshots it took and the steps it ran. We use these records to show you what happened and to improve SpecHawk's agents. Your company's records improve the agent that works on your company's app. Improvements we share across customers come only from SpecHawk's own test apps and from customers who agree in writing to share their records. Improving an agent means changing its instructions, examples and what it remembers about your app, not training an AI model.
How a customer agrees in writing. We don't offer this to customers yet, so no customer's records are shared for now. When we do, an admin of your company will agree for it by ticking "Use our maps to improve SpecHawk's playbooks for every customer" in Settings, beside these words: "Playbooks are how we handle common things like date pickers, pop-ups and sign-in codes, and every SpecHawk customer's maps use them. With this on, your maps can help improve them, and each of your apps keeps its own notes for its next map. Turning it off stops both from the next map on, and deletes nothing. This is on for every workspace unless an admin turns it off here. We keep a record of each change: who made it and when, with their email address." We'll keep a record of each change: who made it, when, their email address, and the words they saw. Turning it off stops new records being shared.
You can ask us to delete your company's session records at any time by emailing us. When your account is deleted, we delete them too.
AI notes on visual checks. We make these only while SpecHawk's own Claude Haiku is switched on. Then, when a visual check finds that a page looks different, we send pictures of that page, cropped to where it changed, and what we measured to Anthropic's Claude Haiku 5.5, which writes a short note about it. When it checks a page for the first time, we send that page's picture for a note too. We never send a picture that shows a saved sign-in detail, or any piece of one. When it's switched off, no AI note is made. A note never approves or rejects a change: a person does. Anthropic keeps those pictures under the Claude API terms above.
AI parts in Agent UI checks. We send a picture from a recording of your agent to our AI only when three things are true: we've switched this on, we've turned it on for your workspace, and an admin of your company has agreed to these words, which we keep with who agreed, their email address and when: "When you turn this on, we send the picture of each answer we record from your agent to our AI (Anthropic's Claude Haiku 5.5), so it can name the parts on screen and suggest what to test. We send only that picture and where each part is, and never a picture that shows a saved password. Checking your agent's screens never sends anything to a model." Our AI here is Anthropic's Claude Haiku 5.5, and Anthropic keeps those pictures under the Claude API terms above.
Notes from earlier maps. When we turn these on for your workspace, after each map we send a stripped list of the map's steps to Anthropic's Claude Haiku 5.5: the paths it visited with any ids taken out, the buttons and links it used, and what happened. The model suggests what to note for the next map of the same environment, and we write the notes in our own words. Apart from the short names of buttons and links, cut to 40 characters, the list and the notes never hold your pages' text, the values on them, anything typed or your saved sign-in details. We keep the step list for up to 7 days, until the notes are written. The notes are kept in that environment's memory store at Anthropic, which only that environment's maps read. A note drops on its own after 60 days, or after 5 maps that don't see it again. In this version they aren't shared with anyone else. An admin can delete any note, or all of them, on the environment's page, or ask us to. We delete them when the environment is deleted and when your account is deleted. Anthropic keeps the list and the notes under the Claude API terms above.
Taste scores. When we turn these on for your workspace, we send pictures of your app's screens, the ones visual checks take, to Anthropic's Claude Haiku 5.5. Each picture is cut to the top of the page, at most three screens tall, and goes with the page's name and the names of the buttons, links and other parts on it. The model scores how each screen looks and suggests up to three fixes. We never send a picture that shows a saved sign-in detail or any piece of one, or one we haven't checked for them, and we take saved sign-in details out of the names before we send them. The scores and suggestions stay in your workspace. Anthropic keeps those pictures and names under the Claude API terms above.
Competitor intelligence. When we turn this on for your workspace, an admin can name two to four competitors. We read only their public pages, the ones anyone can open without signing in, and only where their robots.txt lets us. We keep each page's address, when we read it, and short quotes from it, not copies of their pages. We send the names of your app's features, from your own latest map, and that public text to Anthropic's Claude Haiku 5.5 to compare them, and we take saved sign-in details out of your feature names before we send them. We never sign in to a competitor's site, and we send it nothing of yours. Anthropic keeps that text under the Claude API terms above.
5. Who we share it with
We share data with service providers that help us run SpecHawk. Where they process personal information or Customer Data for us, they are bound by contracts that limit how they can use it.
- Amazon Web Services, for hosting, storage, databases and receiving email, in the United States.
- Cloudflare, which carries traffic to our site and service and protects it from attacks.
- Resend, which sends our emails, such as your welcome email, with your name and email address.
- Google, if you choose to sign in with your Google account.
- OpenRouter, which routes AI requests on SpecHawk's own key to the model your workspace picks.
- Anthropic, for the Claude API and Claude Managed Agents, which run AI features: mapping, test writing, and visual and conversation review.
- Other AI model providers, if your workspace adds its own key, as described above.
- Tools you choose to connect, such as Atlassian Jira and GitHub.
- A proxy service you configure for an environment, if you add one. SpecHawk's browser traffic for that environment then goes through it.
We also send package names and versions, with the package registry they come from (npm), and nothing else, to OSV, the public open-source vulnerability database, to look up known advisories.
We may disclose information if the law requires it, to protect the rights, property or safety of SpecHawk, our users or others, or as part of a merger, acquisition or sale of assets, in which case this policy continues to apply to it.
6. Cookies
We set these cookies:
- One keeps you signed in. Scripts on the page cannot read it, it is only sent over HTTPS, and it expires after 7 days.
- While you sign in with Google, or connect GitHub, one holds a random value, so only the browser that started that step can finish it. It's deleted when that step is done, or 10 minutes after we set it.
- One holds your browser's time zone, like Europe/London, so the dashboard's chart shows your days. It lasts a year.
We also save a few things in your browser's own storage: preferences such as light or dark mode and your list filters, and your recent chats with the Hawk. We do not use analytics or advertising cookies.
7. How long we keep data
- Account data: while your account is open.
- Customer Data: while your workspace is active. When a workspace closes, we delete its Customer Data from our active systems within 30 days of a request, unless your agreement with us says otherwise. Copies in backups are overwritten in the normal course.
- Session records: while your account is open, unless you ask us to delete them sooner. When your account is deleted, we delete them.
- Emailed sign-in codes: 24 hours.
- Logs and usage data: only as long as we need them for security and to run the service.
8. Security
We use HTTPS everywhere, encrypt sign-in secrets at rest, limit access to people who need it, and serve screenshots through signed links that expire within an hour. No system is perfectly secure. If you find a security issue, email us at the address below and we will look into it quickly.
9. Your choices and rights
You can ask us to access, correct, export or delete your personal information, and to delete your company's AI session records at any time, and you can opt out of marketing emails using the link in any of them. Depending on where you live, for example in the European Economic Area, the United Kingdom or California, you may have further rights, such as objecting to or restricting certain processing, and the right to complain to your local data protection authority. We will not treat you differently for using these rights.
If your request is about Customer Data that your company controls, please contact your company's SpecHawk admin. We will help them respond.
Where EU or UK law applies, we rely on these legal bases: performing our contract with you or your company, our legitimate interest in running and improving a secure service, your consent where we ask for it, and compliance with the law.
10. Data transfers
SpecHawk is based in the United States and stores data there. If you use SpecHawk from outside the United States, your information will be transferred to and processed in the United States, with appropriate safeguards where the law requires them.
11. Children
SpecHawk is for businesses and is not meant for anyone under 16. We do not knowingly collect personal information from children.
12. Changes to this policy
When we change this policy, we will update the date at the top. If a change is significant, we will tell you by email or in the product before it takes effect.
13. Contact
Questions or requests: [email protected]