Collabysis uses local browser and Windows workflows, with optional network actions chosen by the researcher. With a trial or paid key installed, the Windows app also requests the signed revocation list from collabysis.com before an analysis or collection, at most once every 24 hours. When a paid or trial subscription key has expired, the next collection or analysis action, including the example the Home screen runs over a recorded collection, also sends the installed license, which holds the purchaser's email address, and a random installation identifier to collabysis.com to renew the key, at most once every 24 hours. With a paid or trial subscription license activated in Settings, the included web searches and AI also send requests to collabysis.com under the conditions that Included searches and AI states, with what each request sends and what collabysis.com keeps. The update check reads collabysis.com/release.json and its signature and installs nothing. Version 3.22.1 checks only when the researcher chooses Check for updates in Settings. Versions after 3.22.1 also check once a day unless the researcher turns off Check for updates once a day in Settings. This page records the public data-flow boundary and links to the fuller policy index.
Each listed network action has a user-selected recipient or a delivery recipient. The browser workbench itself starts with no dataset.
| Feature | Recipient | Fields or data category | Current boundary |
|---|---|---|---|
| Browser CSV analysis | The browser on this device | CSV rows and cells, analysis choices, results, and an optional local session. | The released workbench has no CSV upload endpoint. It does not collect public sources or offer hosted AI. |
| Website delivery | Website host and network providers | Standard request and connection metadata needed to deliver and protect a page. | The released pages contain no advertising or analytics scripts. Provider processing terms need review. |
| Windows collection | The public site, search service, or provider API selected by the researcher | Query terms, target URLs, request metadata, and a researcher-owned credential when that provider path uses one. | Recipients and exact fields depend on the selected source and provider. Source terms remain the researcher's responsibility. Importing a caption, transcript or media file sends nothing. Where transcription is available, it runs on this computer, and the app downloads no model; the installed build does not include transcription in this release. Section 7 of the test guide names the hosts that the researcher-run model download contacts. |
| Find sources and keyless web search | api.gdeltproject.org (GDELT), api.wikimedia.org (Wikimedia) and api.openalex.org (OpenAlex), or first a SearXNG instance that you set. api.github.com, hn.algolia.com and gamma-api.polymarket.com only when the researcher adds GitHub, Hacker News or Polymarket to the plan. | The approved plan's search words, sent only after Approve and find sources, and the Collabysis User-Agent. GDELT, Wikimedia and OpenAlex receive the same search words only. No account, key or email address is sent. | Keyless Web search sends the question keywords to api.gdeltproject.org (GDELT), api.wikimedia.org (Wikimedia) and api.openalex.org (OpenAlex). When you set a SearXNG instance, it is asked first, and the three hosts receive the keywords only when it returns nothing. A source you skip in the plan receives nothing. Collabysis sends no request to DuckDuckGo. None of these hosts is an AI service. Without a web search key, a plan names no Web research, Find sources sends no Web search request, and the Web search card is held for the run. A web search key counts when it is saved in Connect your sources or set in the environment that starts Collabysis. Web pages topic discovery uses the same keyless search, and so does Web research when a web search key is saved and the keyless backend is chosen. In each Find sources, the app reads the robots.txt of each of these hosts before its first request to that host. A host receives no search request when its robots.txt disallows the request, when it refuses the robots.txt request (HTTP 401 or 403), when it asks for a slower pace (HTTP 429), when the robots.txt request fails with a server error, or when the host cannot be reached. The Sources step then names the source with the reason. A host with no robots.txt (HTTP 404, 410 or another 4xx answer) is searched. The app honors Retry-After and refusals. With a paid or trial subscription license, the auto Web search backend and no web search key, a plan also names Web research, and its searches go through collabysis.com; Included searches and AI describes those requests. |
| Start collection from a question's plan | For Web research, which a plan names only when a web search key is saved, the search service the researcher chooses or, with the keyless backend, the hosts of the Find sources row, then the website of each page its results link to, such as doi.org, openalex.org or en.wikipedia.org. With Follow relevant page links checked, as it is by default, Web research also reads pages those pages link to on the same websites and their subdomains, including next pages and feeds. With it unchecked, Web research reads only the result pages. These hosts depend on the results and are not known in advance. For Public data, when the researcher keeps it on the Sources step, api.us.socrata.com for portal search (data.gov search through catalog-old.data.gov is held in this release and sends no request), then the host of the dataset the researcher chooses or, with none chosen, of the portal search's top match, such as a city data portal; or the host of a faster connector the researcher chooses (earthquake.usgs.gov; geocoding-api.open-meteo.com, archive-api.open-meteo.com and api.open-meteo.com; services3.arcgis.com). Public data portal search also contacts www.socrata.com, the host to which api.us.socrata.com redirects its robots.txt request. For the Hacker News source, hacker-news.firebaseio.com for the items of the discussion the researcher names. For each keyless source the plan keeps: api.bsky.app (Bluesky), mastodon.social (Mastodon) and api.wikimedia.org (Wikimedia) in both plans, and api.openalex.org (OpenAlex) and api.crossref.org (Crossref) in a research plan. api.gdeltproject.org (GDELT) when the researcher adds GDELT to the plan, and on each Web research search that uses the keyless backend, unless the researcher's SearXNG instance answers. With a YouTube Data API key saved, the plan also lists YouTube, which sends the search words and your key to www.googleapis.com (YouTube Data API) and reads no robots.txt. With a paid or trial subscription license, the auto Web search backend and no web search key, a plan also names Web research, and its searches go to collabysis.com, which forwards them to api.search.brave.com; Included searches and AI describes those requests. | The search hosts receive two kinds of search. Each keyless source card sends the approved plan's first content phrase, at most four words, to its host: api.bsky.app, api.wikimedia.org, api.openalex.org and api.crossref.org, and api.gdeltproject.org when GDELT is in the plan. The Mastodon card sends mastodon.social up to two tags derived from that phrase, and each tag is read as its own tag timeline; with replies selected, the two tags are sent together as one post text search. A keyless Web research search sends up to eight keywords of the question to api.gdeltproject.org, api.wikimedia.org and api.openalex.org. Every search host receives the Collabysis User-Agent. A result page, a linked page or a dataset host receives a request for that page or dataset with the Collabysis User-Agent; the app adds no question text to that request. No keyless source receives an account, key or email address. A saved YouTube key goes only to www.googleapis.com. | None of these hosts is an AI service. A skipped source is not contacted. Web research reads at most the number of pages set in Pages to extract (20 by default). Unchecking Extract the pages found by web search stops all page reading. Each keyless source reads its host's robots.txt before its first request. Web research reads the robots.txt of each search host before it searches it, and the robots.txt of each website before the first page it reads there. One Start can read the same host's robots.txt more than once. When GitHub, Hacker News, Polymarket or a Public data host redirects its robots.txt request, the app follows the redirect at most five times and sends the target host only the redirected robots.txt request, with no question text. The app honors Retry-After and refusals, and paces GDELT, Wikimedia, OpenAlex and Crossref per host across every run on the computer. Let Collabysis AI choose searches and pages (beta), a switch for Web research on the Sources step, is off by default. With it on, Collabysis AI proposes the searches and chooses the result pages to read inside the approved plan, and the app stops the run at the budget of searches, pages and AI calls shown before Start: at most 15 searches, the number set in Pages to extract and 20 AI calls. With the switch on, each search goes to the keyless hosts of the Find sources row, whatever web search key or license is set. Collabysis AI receives the question, the problem statement, the construct labels and the offered source names and, to choose pages, the titles, snippets and hosts of up to 20 search results or linked pages per choice, each title cut to 200 characters and each snippet to 500; it never receives a page body, a dataset row or a key. With the switch on, the pages read depend on the search results and are not known in advance. The same robots.txt, Retry-After and refusal rules apply to every search and page. AI routes and hosts names the hosts that receive the AI requests. |
| Wikipedia depth | wikimedia.org (daily pageviews) and api.wikimedia.org (revision lists); en.wikipedia.org for its robots.txt only. | Each chosen article's title and the Collabysis User-Agent. No account, key or email address is sent. | For Wikipedia depth, the app sends each chosen article's title to wikimedia.org (daily pageviews) and to api.wikimedia.org (revision lists), after requesting each host's robots.txt. Both hosts redirect that robots.txt request; the app does not follow those redirects and contacts no other host for them. Before the first article it also reads the robots.txt of en.wikipedia.org. The app never requests the /w/ or /api/ paths that en.wikipedia.org's robots.txt disallows, skips any article whose Wikipedia address that file disallows, and the robots.txt setting does not change this. |
| Download media | The host of each media file in the collected records. These hosts are not known in advance. | A request for the media file and its host's robots.txt, with the user agent Collabysis-Media/1.0. No question text, account, key or email address is sent. | Download media requests each media file's host after reading robots.txt at most once per origin per run (scheme, host and port), one request per second per host, honors Retry-After, and sends a host nothing more after a refusal. These hosts come from the collected records and are not known in advance; YouTube pages and videos are never requested. |
| Publisher caption | The host of the caption URL the researcher enters. These hosts are not known in advance. | A request for the caption file and its host's robots.txt, with the user agent Collabysis/3 publisher-caption. No question text, account, key or email address is sent. | Applying a publisher caption reads the robots.txt of the caption URL's host, then requests the caption file from that host; YouTube hosts are refused, and the robots.txt setting does not change this. |
| robots.txt decisions, declaration and export | This computer | The robots.txt decision per host, the first-run declaration and the pseudonym key. | Each collection receipt records, per host, the robots.txt answer the app saw, the rule it matched and the decision it took. Hosts whose robots.txt was not requested before the run stopped are stated as not checked. For X, YouTube (official API), Instagram and TikTok, which use a provider API, no robots.txt is read, and the export says so. With a Saudi Arabia or United States declaration, a Disallow for all crawlers (User-agent: *) does not stop a request by default. You can turn robots.txt honouring on for each source or site in Settings. With an EU or EEA declaration, an Other declaration or no declaration, Collabysis honours robots.txt, and Settings offers no robots.txt setting. The robots.txt setting covers only Hacker News, the open data sources, data file links, the keyless research searches other than Wikipedia, Web pages, the pages that Web research extracts and Download media. It never applies when a robots.txt group names Collabysis or when a host refuses or rate-limits the request, and it has no effect on Find sources, Mastodon, Bluesky or Wikipedia depth. The robots.txt setting does not apply to Wikipedia or the other Wikimedia sites; Collabysis always honours their robots.txt. A change to a source's robots.txt setting applies from the next request. A run records the declaration it started with; if you save the declaration again during a run, robots.txt is honoured for the rest of that run. Web pages requests to a site where robots.txt is not honoured identify Collabysis as "Collabysis-Research/1.0 (+public-data collection)", without the clause "respects robots.txt". A Disallow in a robots.txt group that names Collabysis always stops the request it covers. A refused, rate-limited, unreadable or unreachable robots.txt request also stops the request. Apart from Wikipedia pages, Wikipedia depth and the own-key routes, a 429 answer ends collection from that site for the rest of the run, and its Retry-After is kept as the site's wait for later runs. Apart from Wikipedia pages, Wikipedia depth and the own-key routes, a 403 that carries a valid Retry-After, or a rate-limit reset that the source reads, is handled as a 429 unless the source also reads a CAPTCHA signal in it; any other 403, and any answer other than a 429 in which the source reads a CAPTCHA signal in the headers, cookies or JSON fields, ends collection from that site until you clear it under Stopped sites in Settings. Wikipedia pages and Wikipedia depth keep their own handling: a 429's Retry-After is kept as the site's wait for later runs, a 429 that asks for at most 60 seconds (Wikipedia pages) or 120 seconds (Wikipedia depth) is waited out once and the request is sent again, any other 429 ends the run, a 403 ends the run, and a 403 or a CAPTCHA answer stores no site stop. A site you enter under Stopped sites receives no request, robots.txt included, until you remove it. A declaration change does not remove it. Each collection receipt records every stop decision with the source that received it. Requests to X, the YouTube Data API and ScrapeCreators use your own key and read no robots.txt. A 403, 429 or CAPTCHA answer on these routes stores no site stop, because such an answer can concern your key or quota. A site you enter under Stopped sites receives no request from these routes either. These routes contact only the host of the provider whose key you save: api.twitter.com (X), www.googleapis.com (YouTube Data API) and api.scrapecreators.com (ScrapeCreators). A paid Web search backend whose key you save receives Web research queries in place of the keyless search, at api.search.brave.com, api.exa.ai, google.serper.dev or api.parallel.ai. On the first run the app asks where you work and your role; the answer is stored on this computer, in each collection receipt and in the method note of a coded export, and is sent nowhere. Export and share default to a copy in which author names and handles are replaced by pseudonyms from a key stored apart from the data on this computer; you can destroy the key. |
| Start collection from Web pages | For Web pages, the website of each starting page. A starting page is a URL you paste under URLs I paste or a page that Topic discovery finds with the keyless search of the Find sources row. These websites depend on your URLs and the search results and are not known in advance. With Follow relevant page links checked, as it is by default, Web pages also reads the pages that those pages link to. With Follow in-scope pagination checked, as it is by default, Web pages also reads next pages. With Collect in-scope feed links checked, which is off by default, Web pages also reads feeds. With Record sitemap locations as proposals checked, which is off by default, Web pages also reads the robots.txt Sitemap lines of each website it reads a page from, including subdomains, and requests that website's /sitemap.xml only when robots.txt names no sitemap. | A starting page, linked page, next page or feed receives a request for that page with the Collabysis User-Agent. The app adds no question text to that request. A sitemap request asks for the sitemap file with the same User-Agent. No account, key or email address is sent. | None of these websites is an AI service. Linked pages, next pages and feeds stay on the starting websites and their subdomains. Uncheck Follow relevant page links, Follow in-scope pagination, Collect in-scope feed links or Record sitemap locations as proposals to stop that request. Choose Topic discovery under Starting pages to request no pasted URL. Maximum pages in this collection and Maximum page requests limit the total. The pages that a sitemap lists are not requested. |
| Optional Windows AI assist | The provider selected by the researcher | Route-specific context. Ask and AI-authored analysis receive the question, schema, and method list; AI-authored analysis may also receive prior method names. With provider opt-in, method selection receives the researcher's methodology text, schema names and roles, fixed method catalog, and explicitly selected design-context fields. After disclosure, Define research variables receives selected source-text segments plus their row IDs and content-state markers, the researcher's field definitions, and a source-snapshot digest. Scoped PDFs also include selected-page scope metadata. Semantic Scout receives bounded excerpts after BYO-key opt-in. | Ask, method selection, and AI-authored analysis do not send dataset rows, cell values, or the full CSV. Define research variables sends selected record context; it does not send unselected rows or the full CSV. Semantic Scout removes record identifiers, URLs, source labels, and other record metadata before its request. Provider retention terms require separate review. |
| Downloads | An external release host | Standard request and connection metadata for an installer or ZIP. | The website redirects artifact downloads to that host. Check the version and SHA-256 value before running a file. |
The Windows app contacts an AI service only through the routes below, each under its listed condition. With no AI key set and every AI option off, no AI service is contacted. With a paid or trial subscription license activated, Ask, the Guide or method selection can also reach an AI service through collabysis.com, under the condition in the Included AI row below. With Let Collabysis AI choose searches and pages (beta) on, Web research can also reach an AI service, under the condition in the Web research row below. The reader service for thin pages is included below: its operator is an AI vendor, and listing a site for it counts as turning on an AI option.
| Route | When it is used | What is sent | Hosts |
|---|---|---|---|
| Collabysis AI | A Collabysis AI key and endpoint are set up, and you use an AI feature. The Guide uses this route only when the endpoint is the Collabysis AI default. | The fields of the feature you use, as listed in the table above. Dashboard planning sends your question, schema names and roles, and the capability catalog after you choose AI planning. The Guide sends your message and the current screen name after you choose to ask Collabysis AI. Collection checklist refinement sends your question after you tick its AI option. | openrouter.ai for the Collabysis AI default; otherwise the host of the configured endpoint. |
| Included AI | A paid or trial subscription license is activated in Settings, no other route in this table applies to the feature, and you use Ask, the Guide or method selection. Included AI covers Ask, the Guide, method selection and, with Let Collabysis AI choose searches and pages (beta) on, two Web research tasks: choosing searches and choosing pages. Dashboard planning, AI-authored analysis and collection checklist refinement do not use included AI. Define research variables and Semantic Scout never use included AI. Choosing your own AI provider or an installed AI command-line tool never uses this route. Included AI starts only when Collabysis turns it on. | Ask sends your question and the column names and roles. The Guide sends your message and the current screen name. Method selection sends your methodology text, the column names and roles, and the design fields you select. Web research with the switch on sends the fields that the Web research row below lists. Each request also carries the installed license, which holds the purchaser's email address, and a random installation identifier. Included searches and AI lists each field. | collabysis.com, which forwards the feature's fields to openrouter.ai. |
| Your own AI key | You save your own AI key in Settings and choose a provider, and you use an AI feature. Auto chooses the host from the key format. Semantic Scout uses only this route. | The fields of the feature you use, as for Collabysis AI. | openrouter.ai, generativelanguage.googleapis.com, or api.openai.com, as chosen. With Custom, the host you enter. Collabysis cannot list a custom host in advance. |
| Provider key in the environment | A conventional provider key variable is set in the environment the app starts with or in its private settings file, no provider is chosen in Settings, and you use an AI feature. Each variable goes only to its own host. | The fields of the feature you use, as for Collabysis AI. | openrouter.ai or api.openai.com |
| Installed AI command-line tool | You choose installed CLI authentication in Settings, and you use an AI feature. | Your question or methodology text, column names and roles, the method catalog, and design fields you selected. No dataset rows are sent. | The host the installed tool uses with its own account. Collabysis does not choose it. |
| AI key check | You save or check an AI key in Settings. | The key, once. No question or dataset content is sent. | The host that matches the key, from the hosts above. |
| Collabysis AI advisory checks | An advisory key is set up for the app, and you turn on the advisory option for one action. | Scout source review sends the question and up to eight bounded sampled passages per call. Data request advice sends the need text and fixed route descriptions. Method guidance sends your research aim and fixed method descriptions. Chart guidance sends the chart request. The citation check sends the claim and the cited source passage. The field check sends the field definition, the proposed value, the evidence quote, and the source excerpt. | api.typesafe.ai |
| Record screening | An advisory key is set, and you confirm the screening prompt. | Your research question or stance subject, and the text of up to 48 records. | api.typesafe.ai |
| Reader service for thin pages | You list a site under Reader service for thin pages in a Web pages plan, and a page from that site returns too little readable text. The list is empty by default. Before any reader request the app reads the service's own robots.txt and sends no page URL when it disallows the app's requests. The app never uses the reader after a site refuses, rate-limits or challenges a request. | The page URL and the standard details of any web request: your network address and a user-agent string that names the software. The app first requests the service's robots.txt, which carries the same standard details and no page URL. No question text, dataset row or record is sent. The service fetches the page from its own network and returns the page text. Each page read this way is marked in its record and in the signed receipt. | r.jina.ai |
| Web research | You run Web research with Let Collabysis AI choose searches and pages (beta) turned on for it on the Sources step. The switch is off by default, and with it off Web research uses no AI. With it on, Web research asks Collabysis AI once to propose the searches and once per round to choose the result pages to read, within the 20 AI calls of the budget shown before Start. It uses the first route that applies: the Collabysis AI key your license carries; then your own AI key or a provider key in the environment; otherwise, with a paid or trial subscription license, included AI. | With the switch off, nothing is sent to an AI service: Web research plans the searches and ranks the results on this computer, and it sends the question keywords only to the public sites and search services in the collection row of the table above. With the switch on, the search proposal sends the question, the problem statement, the construct labels and the offered source names, and each page choice sends the question and the titles, snippets and hosts of up to 20 search results or linked pages, each title cut to 200 characters and each snippet to 500. No page body, dataset row or key is sent. With the switch on, the searches go only to the keyless hosts of the Find sources row. With the switch off and included web search, the search text goes to collabysis.com, which forwards it to the web search provider at api.search.brave.com. | None with the switch off. With it on, collabysis.com, which forwards the request to openrouter.ai, on the included route; with the Collabysis AI key your license carries, the endpoint set with that key, from the Collabysis AI row; otherwise the host of the AI key in use, from the Your own AI key and Provider key in the environment rows. |
With a paid or trial subscription license activated in Settings, the Windows app can use the web searches and AI included with the license. Each such request goes to collabysis.com, which sends the search text or the AI task's inputs on to one provider and returns the answer. Included web search runs only with the auto Web search backend and no web search key. Your own web search key and the keyless Web search backend never use it. Included AI covers Ask, the Guide, method selection and, with Let Collabysis AI choose searches and pages (beta) on, two Web research tasks: choosing searches and choosing pages. Dashboard planning, AI-authored analysis and collection checklist refinement do not use included AI. Define research variables and Semantic Scout never use included AI. Included AI runs only when no own AI route applies. Choosing your own AI provider or an installed AI command-line tool in Settings never uses it. Included AI starts only when Collabysis turns it on; until then, collabysis.com refuses included AI requests. Included web search is also off until Collabysis turns it on; until then, collabysis.com refuses included search requests, and your own web search key keeps working.
| Request | What the app sends to collabysis.com | What collabysis.com sends on, and to which host |
|---|---|---|
| Included web search | The search text; the installed license, which holds the purchaser's email address; a random installation identifier; and a random run identifier. | The search text, the number of results requested and an optional date range go to the Brave Search web search provider at api.search.brave.com. |
| Included AI | The AI task's inputs; the installed license, which holds the purchaser's email address; a random installation identifier; and, for a run, a random run identifier. Ask sends your question and the column names and roles. The Guide sends your message and the current screen name. Method selection sends your methodology text, the column names and roles, and the design fields you select. Web research with Let Collabysis AI choose searches and pages (beta) on sends the question, the problem statement, the construct labels and the offered source names for the search proposal, and the question and the titles, snippets and hosts of up to 20 search results or linked pages for each page choice. | The AI task's inputs, with an instruction that collabysis.com holds, go to the OpenRouter AI provider at openrouter.ai. |
| Included search count before a run | The installed license and a random installation identifier. No question text is sent. With Web research selected, the auto Web search backend and no web search key, the collection screen sends this request on its own before you start. With Web research selected, an approved plan and included AI as the AI route, the collection screen can also send it on its own to show the budget line of Let Collabysis AI choose searches and pages (beta), with the switch on or off. | Nothing. collabysis.com returns this month's searches and AI tokens used and left. |
collabysis.com sends the license and the identifiers to no provider. The search text and the AI inputs pass through collabysis.com and are discarded when the answer is sent. For these requests, collabysis.com does not store the search text, the search results, the AI inputs, the AI replies, the email address, the IP address in any form, the installation identifier or a provider key. Provider retention terms require separate review.
| Record | Fields | Kept |
|---|---|---|
| Monthly use per license | License key id, month, searches, AI tokens and the time of the last update. | 13 months: the current month and the 12 months before it, then removed at the next included request. |
| Search use per run | License key id, run id, month, searches and the time of the last update. Only included searches write this record; an included AI request stores no run id. | 35 days after its last update, then removed at the next included request. |
| Request-rate windows | License key id, window kind (minute or day), window start and the number of calls. Every included request counts here, including the search count read before a run. | 2 days after the window start, then removed at the next included request. |
| Monthly totals for all licenses | Month, searches, AI tokens and the time of the last update. This record holds no license key id. | 13 months: the current month and the 12 months before it, then removed at the next included request. |
| Use alerts | Cap kind (search or AI), month, threshold percent and the alert time. This record holds no license key id. | 13 months: the current month and the 12 months before it, then removed at the next included request. |
collabysis.com deletes expired rows when it handles an included search or AI request, or a search count read, that passes the license checks. No scheduled job deletes them. When no included request arrives, a row can remain past these periods until the next such request. The license key id is the opaque identifier in the signed license. The billing records below link it to the Polar license key and to installation identifiers, so these counts are not anonymous. While included web search is off, collabysis.com refuses an included search request before it counts or deletes anything.
The billing service at collabysis.com keeps the records below to activate, renew, limit and revoke licenses. Polar is the payment provider that issues the license key. These records have no column for an email address, a raw Polar license key or a request body. Collabysis has not set a deletion period for most of them; the review status above applies.
| Record | Fields | Kept |
|---|---|---|
| License keys | Polar license key id, license key id, state, and the times of creation, last update and revocation. | No deletion period is set. |
| Device activations | Polar license key id, installation identifier, device label, state, a one-time claim token, Polar activation id, a claim expiry time, and the times of creation and last update. | No deletion period is set. |
| Device bindings | License key id, Polar license key id, installation identifier, Polar activation id, device label, and the times of creation and last update. | Binding rows are never deleted. A replaced device keeps its row and is refused for included searches and AI. |
| Subscription periods | License key id, Polar license key id, Polar benefit grant id, Polar subscription id, Polar order id for a paid period, period kind (paid or trial), term end, state, and the times of the provider's last change and the last update. | No deletion period is set. |
| Trial claims | Polar customer id, Polar subscription id, license key id, trial end and the time of creation. | No deletion period is set. |
| Payment event receipts | Polar delivery id, API version, event type, the id of the changed object, the provider's change time, the time of receipt and the outcome. | No deletion period is set. |
| Revocations | License key id, reason, the Polar order id that led to the revocation, and the times of revocation and creation. | No deletion period is set. |
| Activation attempt limits | A keyed hash of the connection's IP address, with no raw IP address; the window start; and the number of attempts. | Deleted at a later activation attempt once the row is older than two limit windows. A window lasts 60 seconds by default. |
The active workbench uses browser memory. Start over clears the active workbench state. Remembering a dataset is opt-in and is described in the workbench as local storage for up to 30 days.
Use Clear stored data or browser storage controls to remove an optional remembered dataset. Browser settings can affect local storage.
Windows runs, raw records, sessions, exports, reproducibility bundles, handoff artifacts, and logs are local or feature-specific artifacts. This public page does not set an automatic deletion schedule for them.
Owner and implementation review are required to publish paths, retention defaults, deletion controls, handoff recipients, and log treatment.
Contact al@collabysis.com for a privacy, accessibility, or security concern. Do not send credentials, private datasets, or active exploit payloads.
The app's Send feedback link opens a message to al@collabysis.com in your own email program. The app sends nothing; the message holds the app version and the current screen.
