Privacy policy
Built by Tradi3. Operated by Vaytric, which is responsible for the data described here. This page covers the hosted site at bountyoperator.com. The open-source command-line kit runs on your own machine.
No ad cookies. No tracking pixels.
No third-party analytics scripts either.
Your files are not stored
Files, instructions, your own AI provider API keys and review results are never written to our database or logs.
Two cookies
Both exist to sign you in. Nothing else is set.
Export or delete
Download your account data or delete the account from the account panel.
Your files and your API key
File selection, the privacy check and prompt export run in your browser.
A hosted review sends the files you selected, your instructions, the model name and your API key through our Cloudflare Worker to the provider you select. The Worker adds the review method of the profile and holds everything in memory for that one request. Your files, your instructions and your key are never written to our database or our logs, and neither is the review that comes back.
The Worker reads the answer to a hosted profile as it passes through, to stop an answer that repeats the method in place of a review. It keeps nothing of what it reads.
The provider you select receives your files under its own terms and keeps them under its own retention policy.
Exporting the prompt of a core profile contacts no provider and sends nothing of yours to our server. You paste it into a model yourself.
Account records
An account is a random identifier. Sign-up asks for no password and no email address.
We store:
- your passkey public keys, with their identifiers, counters, transport hints, labels and creation times
- a hash of your recovery code
- hashes of your session tokens, each with a CSRF token, its expiry and the time you last signed in
- hashes of your connection tokens, with the label you gave each one and its creation, expiry and last-use times
- the billing references listed under Payments
Passkey private keys and device biometrics stay on your device.
Cookies and browser storage
Two necessary cookies: a session cookie (up to 30 days) and a five-minute sign-in cookie.
__Host-bounty-sessionkeeps you signed in. It lasts up to 30 days.__Host-bounty-challengeties a passkey prompt to the request that started it. It lasts five minutes.
Both are HttpOnly, Secure and SameSite=Lax.
The site’s scripts keep the following in your browser. None of it is a cookie, and our server never reads it. “This tab” is the tab’s session storage: closing the tab removes it. “This browser” is local storage, or the history database, and stays until you clear it.
| Name | Kept in | What it holds |
|---|---|---|
bo-motion | Kept inThis browser | What it holdsWhether you pressed Pause motion or Play motion, when that differs from what your system asks for. |
bo-theme | Kept inThis browser | What it holdsNothing now. An earlier version of the site kept your colour theme here; the site removes it when a page loads. |
bo:history, bo-history | Kept inThis browser | What it holdsReview history, when you turn it on: the packet, manifest and text of each finished review, with its profile, model, verdict and time. |
bo:known | Kept inThis browser | What it holdsThe value 1 once you have signed in here, so the sign-in dialog offers Sign in first and the header button reads Sign in rather than Sign up. Deleting the account removes it. |
bo-bench-hidden | Kept inThis browser | What it holdsThe leaderboard columns you hid on the benchmark page. |
bo:board-seen | Kept inThis browser | What it holdsThe date of the bounty board you last looked at, so the next visit can mark what is new since. It is never sent anywhere. |
bo:workbench | Kept inThis tab | What it holdsThe files you loaded, your focus and context, the review type, provider and model you chose, the step you are on and the last result. |
bo:workbench:view, bo:workbench:stash | Kept inThis tab | What it holdsWhether an example is open, how the review runs, which provider’s key field is open and the privacy warnings you accepted; and your own work while an example is shown. |
bo:handoff | Kept inThis tab | What it holdsA draft or files a free tool, a template page or the guide passes to the workbench. The workbench removes it when it reads it. |
bo:openrouter:kept | Kept inThis tab | What it holdsYour OpenRouter key, only when you tick “Keep the key for this tab session” as you connect OpenRouter. |
bo:openrouter:pkce | Kept inThis tab | What it holdsThe one-time code of an OpenRouter connection while it completes. |
bo:gauntlet:run | Kept inThis tab | What it holdsThe answers of the Gauntlet stages that have finished, so a run picks up where it stopped. No key and no source file. |
bo:panel:seats, bo:panel:run | Kept inThis tab | What it holdsThe providers and models of your Panel seats, and the reviews the Panel has finished. No key and no source file. |
bo:return | Kept inThis tab | What it holdsWhere to put you back on the page after Stripe Checkout, for up to two hours. |
History is optional and is saved only in this browser. It never includes source files, API keys or GitHub tokens, it is never sent to our server, and one button clears it. A GitHub token you type for an import is held in the tab’s memory and is not stored.
Usage records and abuse limits
For each hosted review we store the account identifier, the status (running, completed or failed), the review profile, the channel (web or MCP) and the timestamps. These rows enforce the daily allowance and the concurrency limit. They are deleted after seven days.
Abuse limits use a keyed hash of your IP address that expires within a day. Expired sessions, sign-in challenges and rate-limit rows are deleted daily.
The Worker keeps no invocation logs. When a request fails, its own log line holds the request path and an error name, and for a failed billing call Stripe’s error type and code. No log line holds an IP address, a request body, file contents, an API key or a Stripe payload.
Cloudflare, which hosts the site, processes each request’s IP address to deliver it.
Page and funnel counters
Page views and named actions are counted in aggregate. We do the counting ourselves and keep one number per day for each of these:
- pages seen: each page reports once that it has loaded and is on screen, unless the browser says automation is driving it
- page views per path, and each day’s total of the views a browser asked for as a page, which the request’s own headers say
- the referring site, from a fixed list: search engines, X, GitHub, Hacker News, Reddit, the bounty platforms, the MCP and skills directories glama.ai, mcpnav.net, mcpservers.org, mcpmarket.com, skillsmp.com and skillsdirectory.com, the Disclosed newsletter, and the assistants ChatGPT, Perplexity, Claude, Copilot and Gemini. Google Search, Google Search Console and Google’s other sites each count apart. Any other site counts as “other”
- sign-ups and sign-ins
- completed reviews per profile, failed reviews per kind of failure, and answers withheld per profile
- daily-limit hits, checkouts started and subscriptions activated
- named actions a page reports from a fixed list: an example loaded, a prompt exported, a reply pasted, a packet saved, a repository imported, the sign-in opened, the upgrade button pressed, a gauntlet or a panel started and finished, a free tool run, a template copied, an MCP command copied, a template draft passed to the review, a Create a free account button pressed, the bounty board searched or filtered
- on the MCP endpoint: sessions opened, tool calls per tool, and calls turned away at the limit. A session is counted under the kind of agent that opened it, from a fixed list: Claude, Codex, Cursor, or “other”. A registry or a monitor is counted apart
A page reports an action by its name and sends nothing with it, and it reports that it was seen the same way. On a free tool, a template page and the MCP page that counter is the only request your action causes: what you paste or drop there never leaves the browser.
Our release checks and our account audit name themselves in each request. Their page views, named actions, MCP sessions and tool calls, sign-ups, sign-ins and checkouts are not counted.
A counter holds a date, a name and a total. It holds no account identifier, no IP address and no cookie value. Counters older than 400 days are deleted.
Cloudflare, GitHub, OpenRouter and MCP
Cloudflare, which hosts the site, processes each request’s IP address and other request metadata to deliver it, under its own policies.
A GitHub import goes from your browser straight to api.github.com, with the link you chose and your read-only token if you supply one. Without a token the files themselves are read from raw.githubusercontent.com, GitHub’s file host, which is sent nothing but the file’s address. The token is held in tab memory. The imported files reach our server only when you run a hosted review.
Connecting OpenRouter happens between your browser and openrouter.ai.
When you press “Check current free models”, your browser requests the public model catalog from openrouter.ai. That request contains no files, API key or cookies. OpenRouter receives ordinary network metadata, including your IP address, under its own policies. The fetched list stays in this page’s memory; only the model you select is kept with the existing workbench settings.
The MCP server handles the file contents your agent passes to a tool. list_profiles, prepare_review and build_packet keep none of it. run_review is a hosted review and is recorded like one. Your agent’s own model provider receives tool results under its own terms.
Connection tokens are shown once and stored as hashes. They expire after 90 days and you can revoke one in the account panel. Recovering an account, or signing out everywhere, revokes all of them.
Optional affiliate links
Novita signup links marked “affiliate link” are optional outbound links. If you follow one, Novita receives the referral and campaign parameters in that link, along with ordinary network metadata such as your IP address and browser information. Novita may use cookies or other provider-side tracking to attribute a signup or purchase under its own policies.
We may earn a commission if you sign up and pay for Novita through that link. Commission does not determine our defaults or benchmark rankings. You can use the direct API-key management link for an existing account without following the affiliate link.
These links add no tracking scripts, pixels or affiliate cookies to Bounty Operator. We do not put your files, prompts, API keys or Bounty Operator account identifier in affiliate URLs. Referral parameters are used for voluntary website navigation only, never in model inference requests. A hosted Novita review follows the same file and API-key path as other providers; Novita processes that content under its own terms.
Payments
Stripe collects your email, card and billing details at checkout. Full card details never reach our application. We send Stripe your account identifier.
Our database keeps the Stripe customer and subscription IDs, the subscription status, the paid-until time, your receipt email, and webhook event IDs for 30 days. Stripe keeps its own financial records.
Model API and prepaid credit
Model API requests pass through Cloudflare to the hosting provider for your selected model, shown in the catalog. We do not log prompts, tool arguments or responses in our application or use your content to train models.
Our self-hosted models run on vLLM on RunPod, with request and response content logging and cross-request prefix caching disabled. Managed models run through Infron or OrcaRouter. Those services and any downstream model provider process your content under their own terms and retention policies; we do not guarantee zero retention for managed models.
We store hashes of API keys, their visible prefixes and labels, expiry and revocation times, spending limits and aggregate spending. We store each request’s model ID, key ID, time, status, input/output token counts, reserved credit and final charge. Request metadata is deleted after 30 days. Prompts, tool arguments and model responses are not written to our database.
We keep the credit ledger, including paid-month grants and reversals, and Stripe checkout/payment references for accounting and disputes. Deleting an account removes its link from those records, but does not erase the financial records.
If you opt in to auto top-ups, Stripe saves your payment method. We keep its Stripe reference, your top-up settings, the time and version of your consent, and payment attempts and outcomes. Full card details stay with Stripe; we retain the method type and masked details so you can identify it. Cloudflare and the selected hosting providers process traffic and infrastructure metadata under their own policies. Model processing may take place outside your country.
Export and deletion
Download your account data or delete your account from the account panel. Cancel an active subscription and settle any API credit, outstanding balance, running request or pending payment first. Contact support to refund unused credit.
Deleting removes the account, its passkeys, sessions, connection tokens, API keys, request metadata and subscription references from our database. Prepaid financial records remain with the account link removed. It cannot be undone.
Questions: support@bountyoperator.com.
Changes
We update this page when what we process changes. The date at the top is the current version. The security page shows how the stored data is protected.