Privacy policy
Last updated: 16 November 2026.
This policy describes how the Tohou Sonar application handles account and workspace information. The current service operator and contact for privacy requests is Ichate, Switzerland: [email protected]. Tohou is not represented here as a registered legal entity.
Representative in the European Union
Tohou is established outside the European Union, and the Reef sensors described below record and publish addresses that can belong to people in it. Under Article 27 of the GDPR we therefore name a representative in the Union, who can be contacted by data subjects and by supervisory authorities on every matter in this policy, alongside or instead of us:
Boobica David
Avenida Cervantes 5
48310 Elantxobe
Biscay, Spain
[email protected]
Naming a representative does not move responsibility off us: Tohou remains the controller for everything described here, and a complaint answered by the representative is still answered on our behalf.
Information processed
Discord sign-in supplies your Discord user ID, display name, email address and avatar as associated with your Discord account. Sonar does not access your server memberships or messages. We store workspace membership, submitted targets, verification challenges, scan results, findings, comments, API-key metadata (name, scopes, workspace and usage times) and posture score snapshots. API secrets are stored as hashes. Saved Discord webhook URLs and authenticated test credentials are encrypted. One latest service screenshot may be stored per observed web host.
Purpose and sharing
We use this information to authenticate users, verify scope, provide scans and reports, enforce limits and investigate misuse. Workspace data is accessible to authorized workspace members and the operator responsible for maintaining the service. We do not sell workspace data or publish it in a global index. That sentence is about workspace data — your assets, scans and findings. It is not about traffic sent to machines we operate ourselves, which we do publish; that is a separate body of data and the sensor section below describes it in full.
Discord processes authentication and receives notification content when you configure a Discord or Slack incoming webhook. Domain and brand lookups send the queried name to public certificate-transparency services. Public Index lookups send the queried domain or address to third-party public record services: public certificate-transparency logs, public web-observation services, registration directories, regional internet registries, geolocation data providers and public routing data. Domain lookups also resolve public DNS records. DNS resolvers and target systems receive the network queries needed for verification and scanning. Hosting and network providers process traffic and stored service data as part of operating the infrastructure.
Workspace scans also query public sources about your own assets: verified domain names go to the responsible domain registry’s RDAP service (registration expiry), lookalike spellings of them go to DNS resolvers, scanned public IP addresses go to Shodan’s keyless InternetDB, and observed software product names and versions go to the US National Vulnerability Database (NVD) and OSV.dev to look up known vulnerabilities.
When an administrator imports a CycloneDX software inventory for a verified asset, Sonar sends versioned package URLs to a public open-source vulnerability database to check for known vulnerabilities. The uploaded document is processed in memory and is not retained; Sonar stores its digest, component count and resulting workspace findings.
ASN coverage requests send the ASN to a public routing-data service and store the returned public BGP prefixes in the requesting workspace. These routes are passive observations, not proof of asset ownership. Written authorization submitted for operator review is encrypted at rest and accessible to the platform operator for that review. It is retained with the workspace authorization record and removed when the prefix or workspace is deleted. Company coverage labels and their links to already verified domains are workspace-private.
The Tohou Index is stateless. Searches and their results are never written to the service database and there is no per-user query history. A lookup record is held only in volatile server memory for up to 15 minutes so repeated identical searches can be answered quickly, and it disappears on restart. A search sends the entered domain or address to the third-party public record services listed above; those services have their own logging and retention practices. Domain lookups additionally query a second public certificate feed, public subdomain feeds, a public web crawl index and a public web archive of historical pages; address lookups query a public routing registry. All of these are third-party services — no query ever sends traffic to the researched target.
A bounded direct observation is available only for targets inside the requesting account's own verified scope: single connection attempts to a short list of common service ports, one HTTP(S) request for status and title, and a TLS certificate read when a web port answers. Tohou never sends probes from its infrastructure to a target the requester has not verified ownership of; every other Index lookup is answered from third-party public records alone. The operator can disable live observation entirely with the INDEX_LIVE environment flag.
Screenshots are captured by loading your verified web service in an isolated headless browser session, exactly as a visitor would. Only the latest screenshot per host is stored, it is visible only to workspace members through an authenticated route, and it is deleted together with the asset.
Index API requests are authenticated with hashed key secrets, counted against the account's daily lookup budget, and never stored as query history beyond the same volatile cache described for the Index.
Brand monitoring sends a verified domain label in a query to a public certificate-transparency service and stores potential certificate name matches inside the requesting workspace; these matches are leads for review, not a conclusion that a site is malicious.
If a workspace administrator enables credential-exposure monitoring for a verified domain, Sonar matches that domain against a public catalog of known breach records and stores only aggregate information in the workspace: breach names, dates, affected-account counts and exposed data categories. No credentials are ever requested, fetched or stored, and no administrator-provided keys are involved.
Attacker telemetry from our sensors
Tohou operates inbound sensors on its own addresses. They accept unsolicited connections, record what the sender tried and close. They never initiate traffic, never scan the sender and never respond with an exploit. The source address of that traffic is the central record, and we publish those addresses through the Reef API, a bulk feed and the Tohou Index. An IP address can be personal data, and this data does not come from the person it is about, so this section is the notice required by Article 14 of the GDPR.
For each observation we process the source address, the time, the destination port and protocol, a bounded and sanitised slice of what was sent, any signature or CVE identifier it matched, and the network (ASN) and country the address is registered to, derived from public routing and geolocation data. Nothing else about the sender is collected.
The people this can concern are senders of unsolicited traffic to systems we operate — in practice mostly automated hosts, many of them compromised machines whose owner does not know they are sending it. We do not profile individuals. We do not join this data with account, workspace or any other identity data we hold. We do not enrich it from commercial identity, email or people-search sources. We do not sell it and we do not give it to data brokers. We do not try to establish who is behind an address, we never describe an address as malicious, and the absence of an address from the data does not mean it is safe.
Login attempts are part of what the sensors see. Attempted usernames are aggregated on the sensor itself and named only once enough unrelated senders have tried the same one, so an account name reaches us as a word many attackers tried, never attached to the address that tried it and never beside a password. We do not publish those lists. Attempted passwords are counted and discarded where they arrive: the sensor keeps no list of them, so no attempted password is transmitted from a sensor, stored in Tohou’s database or published, in any form, hashed or otherwise. The ingest API has no field that could carry one and refuses a whole submission if it carries a single field the API does not expect, so there is no route by which a password could arrive even from a sensor that had been tampered with.
Our lawful basis is Article 6(1)(f), legitimate interests: running and defending a security service, and telling others which weaknesses are being attacked and from where. Recital 49 names network and information security as such an interest. The balancing, briefly. The purpose is to show which vulnerabilities are being exploited right now, and that a given address is taking part. It is necessary because nothing short of observation establishes that exploitation is real and current. The address is the indicator because it is the only identifier a connection carries and the only one a defender can act on. Against that we leave out what would make this more intrusive: request payloads are never published, credential material is never kept, no identity lookup is ever performed, addresses on mobile and shared networks are held to a lower confidence tier because one of them can carry thousands of unrelated people, and an address below our confidence threshold is not published at all. The sender also chose to contact a machine of ours that they had no reason and no permission to contact.
The API, the feed and the Index are reachable worldwide, so this data is accessible outside the EEA by design. Publication is the point of it. The service is operated from Switzerland, which the European Commission recognises as providing adequate protection, but we are not claiming a transfer mechanism for the published output: a public feed is public.
Retention is per record type and runs automatically. Raw sensor batches are kept 7 days; individual events 30 days; session identities and fingerprint clusters 30 days; exploitation source records 90 days. A published reputation entry carries its own expiry, decays once the activity stops, and is deleted no later than 180 days after the address was first seen whatever it did in between. Per-window sensor aggregates — counts, targeted ports and the k-anonymous account-name lists, none of which contains an address — are kept 180 days, and aggregate daily counters, which also contain no address, are kept longer. Delisting requests and the sensor audit trail are kept three years, because a decision to publish or remove someone’s address has to stay reviewable; the audit trail is keyed on a hash of the address rather than the address itself, so accountability survives the erasure of the address.
You can ask what we hold about an address you control, object to the processing under Article 21, and request erasure under Article 17. Write to [email protected] about a published address and to [email protected] for anything else. We will not ask you to prove who you are, only that you control the address; identifying the person would be a worse outcome than the one you came to fix.
Delisting a published address
Write to [email protected] with the address or range and what has been done about the traffic. For a single address we issue a one-time token to publish as the entire body of /.well-known/tohou-delist on that address, and we read it back once from our side. For a range, we use the abuse contact registered for it at the regional internet registry and the operator reviews the request by hand. Proof of control is required so that nobody can delist somebody else’s address. Your contact details are encrypted at rest and are never published.
We acknowledge a request within five working days and decide within thirty days of receiving a complete one. If we grant it, the address stops appearing in the Reef API, the bulk feed and the Index within one publication cycle, and a lookup for it returns exactly what a lookup for an address we have never seen returns. Suppression lasts 180 days; it is not a permanent allowlist. If we refuse, we say why.
Our sensors keep recording the address either way. If hostile traffic from it resumes, that traffic will be recorded and published again. We will not promise permanent removal while we go on observing, because that would be dishonest: delisting ends the consequences of what has already happened, not of what happens next.
AI assistance
When you request a finding explanation, Sonar sends that finding's title, severity, evidence, remediation text and hostname to the DeepSeek API (Hangzhou DeepSeek Artificial Intelligence Co., Ltd.) for processing. DeepSeek states that API data is processed and stored in mainland China and retained under its own policy. Do not include secrets or personal data in target names, evidence or comments. Explanation requests are made only when a workspace member explicitly asks for one.
After each domain scan, Sonar also sends the same provider a compact map of that scope to produce the AI surface summary shown on the Surface map: hostnames, resolved IP addresses, the detected CDN provider, open port numbers, and the titles and severities of open findings. Evidence, response bodies, credentials and secrets are never included. The platform operator can turn this off independently of explanations. For high-severity or known-exploited CVE findings, Sonar also asks the provider for a short impact note using only the public CVE record (the CVE identifier, its public NVD or OSV advisory description, severity and known-exploited status). No hostname, workspace, evidence or other customer data is included, and one note is shared by every finding for that CVE. See the DeepSeek Privacy Policy and Open Platform Terms.
Cookies and local storage
Sonar uses an essential session cookie lasting up to 30 days, shared across the Tohou subdomains you are signed in to (console and index), plus a short-lived OAuth state cookie. Theme preference is saved in local storage. The application does not include advertising trackers or nonessential analytics cookies.
Retention
Current asset state and findings remain while the workspace exists. Changed observations are retained for up to 90 days, with at most 30 historical states per asset. Notifications and completed scan records are retained for 30 days. Screenshots are latest-only per host and removed with the asset. Posture score snapshots are daily aggregates retained while the workspace exists. Cleanup requires an operational worker.
Deletion and exports
You can export workspace findings as CSV or export assets and findings as JSON. Account deletion is scheduled for 24 hours after confirmation. At the deadline, the worker removes the account and all owned workspaces; downtime can delay execution. Personal comments are removed during account purging. Other organizations retain their own asset and finding data. Copies you or recipients export to external systems are outside Sonar's control.
Infrastructure snapshots and independent operator backups may retain data separately. The operator must disclose any backup retention before enabling backups for customer data and must prevent restoration from reversing completed deletion requests.
Your choices
You can revoke API keys, remove webhooks, manage workspace access and request account deletion. Contact [email protected] to request access, correction, or information about processing and any rights available under applicable law.