Privacy notice

Public browsing and MCP requests

The public directory and stateless HTTP MCP process the search terms, filters, software identifiers and tool arguments you send to return results. They do not create an application session, issue a tracking cookie or accept public feedback writes. They do not record the legacy session-based discovery-observation store, but recorded requests can contribute to the separate memory-only aggregate counters described below. The application does not save raw search text as directory analytics.

This does not mean that no personal data is processed. The internet-facing server routing layer records source IP addresses, requested TLS hostnames (SNI), connection times and byte counts for operations and security. Warning/error diagnostics can include request paths or query strings. The application also uses the backend connection address in in-memory request-limit state; behind the proxy this can be the proxy address. Avoid private data in search terms or URLs.

Session-free first-party request counts

Separate in-process counters aggregate instrumented website page, search, detail and comparison events and completed MCP tools/call outcomes. Allowlisted outbound redirects can also count as website outbound events; successful feedback submissions and the legacy click-recording endpoint are limited to the controlled trial. MCP counters use fixed directory tool names, a generic category for unknown tools, outcome categories and directory-processing latency ranges. Protocol initialization, tool listing and ping are not tool calls. These are explicitly recorded logical events, not every network request, asset, person or vendor operation; retries can count again.

The request User-Agent can be inspected transiently for a coarse suspected-bot classification without retaining its original text. A declared-test category requires trusted process configuration; unclassified events can still include bots, tests and people. This counter module does not retain raw IPs, queries, URLs, tool arguments, User-Agents, error text, sessions or fingerprints. Public request summaries show only UTC-day/channel/action unclassified buckets with at least five events. Declared tests and suspected bots are omitted, but unknown ones can remain. Hidden-bucket totals, detailed outcomes and latency ranges are not publicly disclosed. The threshold is not an anonymity guarantee.

Request counters stay in memory for up to 30 UTC days, including the incomplete current day. Expired buckets are purged on reads/writes and by a non-blocking one-minute timer. Restarting the process clears the counters. Earlier or uninstrumented traffic is unknown, not zero; historical logs are not backfilled. The /dev/traffic page is a local-development loopback demonstration and is unavailable in public and SSH-trial modes. No production internal metrics socket is configured or enabled, and no production administrator authentication is claimed.

Software-level directory interest

A separate memory-only interest module records registered catalog software IDs with MCP interfaces, UTC dates, coarse traffic classes and three counts: search_appearance, detail and integration_redirect. A search appearance means a software item appeared in returned results, not that the user searched specifically for it; a single instrumented result list can record up to 20 eligible software appearances. Detail means a directory information read. These counts are separate from request totals and must not be added to them. No raw query, IP address, session, User-Agent, URL, fingerprint or event-level timestamp is stored by this module.

An integration_redirect records a GET to an allowlisted directory jump link that returns HTTP 302 to that software's recorded homepage, official documentation or pricing link. The /out/{id}/{hash} route rejects arbitrary destinations; it does not fetch the vendor page, proxy business calls or establish that the vendor received the request or executed work. Links mapped in the interactive detail view can be observed this way. Direct external links in server-rendered pages are not all covered, so these are not all site clicks or all vendor visits.

Public interest rankings use a selected seven- or 30-day UTC window. A software entry appears only when its three unclassified event counts sum to at least five in that window; this differs from the per-day/action request threshold and the legacy three-session publication rule. The three components are displayed separately, without a weighted quality score. Bookmark counts are null because bookmarking is not implemented. External rankings are shown separately. Declared tests and suspected bots are excluded, but unclassified events may still include repeats, tests and bots. This is not a count of people, independent sessions, vendor executions or success, and the threshold does not guarantee anonymity. Interest buckets are kept only in process memory, purged on reads/writes and a one-minute timer within a 30-UTC-day window, reset on restart and never backfilled from historical logs.

Controlled trial and optional submissions

The separate local/SSH trial uses a one-hour session identifier and a CSRF token. Its application cookie is described on the cookie page. Optional agent reporting requires explicit opt-in and a scope valid for up to 15 minutes and 20 report attempts; it can be revoked in that session. These are trial controls, not verified identities or production account authentication.

When you choose to contribute, stored review data can include a software/interface reference, version, task, capability, rating, title, body, usage context, timestamps and a session digest. Reports can include a review reference, reason, details and moderation outcome. Voluntary usage records include software/interface and client versions, scene, capability, occurrence time, separate discovery/vendor outcomes, error category, evidence level, private notes, consent metadata and deduplication digests. No vendor credential or raw execution log is requested. Input checks reduce sensitive submissions but cannot guarantee detection of all personal data or secrets.

Authorized maintainers review submissions. Approved reviews may be shown with their self-report status. Usage records are published only as permitted aggregate groups; raw notes, session digests and report details are not public. The public service has separate state and does not automatically copy private trial submissions.

Discovery observations and their limits

In the controlled trial, first-party search, detail, integration-guide and explicit outbound-click events can be counted by UTC day and software. Stored buckets contain event counts and daily salted session digests, not raw queries, IP addresses, project content or vendor execution logs. A bucket needs at least three sessions before publication; different sessions do not establish different people.

Voluntary usage summaries use approved records from the latest 30 days and groups with at least three separate sessions. Self-reports and client observations are not platform tests. Search, clicks and integration-guide views do not prove a vendor was contacted, executed work or succeeded. No global execution total or overall success rate is produced.

Storage, access and third parties

The current server and application JSON stores are on Tencent Cloud Lighthouse infrastructure in Seoul, South Korea. The operator is a company in China and authorized administration can involve access from outside the server region. This location notice is not a claim that cross-border legal requirements have all been assessed or satisfied. Do not send sensitive personal information.

Restricted server permissions and separate public/controlled stores limit access, but do not establish absolute secrecy or encryption of every stored item. Technical administrators can access server data when necessary for operation, support or incident handling. Statutory disclosure obligations may apply.

There are currently no external advertising or analytics scripts in the site. Vendor links navigate directly or through an allowlisted directory redirect; your browser then contacts services with their own terms and privacy policies. CapaLens does not fetch the vendor page for you or conduct its business operations. Sending email uses your mail provider and the operator's email service, which process sender details and message content under their own arrangements. The site does not send email automatically when you open a mail link.

Retention and deletion

Trial sessions become invalid after one hour; agent grants expire sooner as described above. Expired in-memory state is cleaned when the relevant code runs or the process restarts. Rate-limit windows are not a promise that IP state is physically deleted at the end of each window.

Legacy controlled-trial session-based discovery buckets use a rolling 30-day retention rule. Cleanup runs on module reads/writes, not a scheduled background eraser; expired files can remain until cleanup is triggered. The public read-only process only filters any expired legacy buckets in memory and does not rewrite a private file. The independent request and software-interest modules instead use the memory-only read/write and minute-timer cleanup described above; neither persists a history across restarts.

Reviews, reports, usage submissions and their moderation records currently have no automatic expiry. They require maintainer review and deletion when no longer needed or when a request is upheld. The usage-record forget action removes the record, notes, digests and related moderation explanations from the live store, leaving only a content-free deletion audit with record identifier and time. It does not automatically erase exports or backups.

Nginx is configured for daily rotation with ten rotated files, compression and exceptions for missing/empty logs. This is not a guaranteed ten-day deletion deadline. No fixed maximum time retention was explicitly configured for the system journal in the reviewed configuration; rotation, storage limits and administration affect actual retention. No uniform deadline for all infrastructure logs, support mail, exports or backups has been verified. These items require manual retention review; we do not promise automatic destruction on a fixed date.

Your choices and requests

You can browse the public catalog without enabling trial reporting. Stop optional submissions or revoke the agent scope to stop new agent reports; revocation does not itself erase already submitted records. You can clear the trial cookie in your browser.

Contact the privacy email below to ask about access, correction, deletion, withdrawing consent, restricting or objecting to processing, or other rights available under applicable law. Include the public page or record/receipt identifier and enough context to locate the issue. Do not send passwords, tokens or identity documents unless a necessary, proportionate verification method has been agreed. We may need to establish your relationship to a record and explain any lawful reason that limits a request.

You may also contact the competent data-protection authority where applicable. We do not claim that a particular foreign authority has approved the service, or that every jurisdiction's requirements are automatically satisfied. This notice will be updated when the implementation or relevant processing changes.

Operator and contact

南京定义未来科技有限公司
Nanjing Define Future Technology Co., Ltd. — reference translation; not a verified registered English name

Customer service and privacy requests: service@deffuture.com