Lockryn

Lockryn — AI protection for web services and servers: AI advises.Policies allow.People decide.

Background illustration: a server log stream. Among ordinary events, failed sign-ins from one address appear; a rule detects password brute force, and a policy allows a temporary 24-hour block.

Lockryn collects telemetry from your servers, detects attacks and outages, explains incidents with AI and takes only the reversible actions your policies allow. Anything critical needs a human's approval.

Get early access Watch an attack's path

MVP in development · early access

03:14the first log line

One attack. Five frames.

SSH password brute force — from the first log line to the person who can cancel the block. Illustration: IP_1 and USER_2 are de-identified placeholders; this is how AI sees the data.

  1. Collect

    A line in the log

    Open-source agents on your servers read logs, metrics and file changes. Passwords, tokens, card numbers and personal data are masked before anything leaves the server, and events land in the event store in the EU, tagged with the tenant.

    auth.log · Failed password · USER_2 ← IP_1

  2. Rule

    Detection

    Detection rules and time-window correlations check the event stream continuously. Dozens of failed sign-ins from one address in a minute is password brute force. Detections of the same kind are merged into one incident with priority P1–P4.

    SSH password brute force → detection → incident

  3. AI triage

    Explanation

    AI receives de-identified context, explains the incident with links to evidence and proposes actions from the catalogue only. It has read-only tools and carries out nothing itself. Without AI, the system runs on rules.

    verdict: attack · evidence: log events

  4. Policy

    Permission and block

    A policy engine checks the proposal: is the action in the catalogue, is the target yours and not on the allow list, which autonomy level is allowed, is the rate limit exceeded. An allowed block is applied by the agent on your server and lifts itself once its time runs out.

    temporary block of IP_1 · L2 · TTL 24 h

  5. Human

    The final say

    People get a short notification in their language: what happened, what has been done and what is proposed. Any action can be cancelled in one click. Details stay in the console: internal addresses and log fragments never go to messengers.

    Telegram notification · “Cancel the block”

One switch — and Lockryn only notifies.

The kill switch puts a tenant into “notify only” mode: Lockryn blocks nothing by itself and only tells people. A cap on automatic blocks per interval works the same way: once it is exceeded, the system switches to this mode by itself.

A viewer cannot turn on the kill switch or cancel actions.

  • Temporary IP blockwas: L2 · automatic, with TTLnow: notify only
  • Block for a medical profilewas: L3 · with human approvalnow: notify only
  • New catalogue actionwas: dry run · “would have done”now: notify only
  • Notifications to peopleL1 · sent

Try it: click or press Space. This is a demo — it switches nothing off.

Features

What the first release (MVP) includes. The set will grow after the pilots.

  1. Availability and health

    Checks from inside and outside, TLS certificates, domains. An agent that goes silent is a signal too.

  2. Attack detection

    Detection rules, correlations and detectors of the agents on your servers. Detections of the same kind are merged into one incident.

  3. AI explanations

    Verdict, priority, evidence and recommended actions in Ukrainian and English. Every conclusion links to evidence.

  4. Safe response

    Temporary IP blocks that lift themselves, dry runs, an allow list for administrators and monitoring, one-click cancel.

  5. Threat knowledge base

    CVE/NVD, CISA KEV, EPSS, OSV, MITRE ATT&CK and CERT-UA are matched with your software, so you only see what concerns you.

  6. Notifications and digests

    Email and Telegram with buttons, a signed webhook, including for the Cifrela bot (planned). The recipient's language and schedule, no internal addresses or log fragments.

  7. Console

    Overview, incidents, services, settings. Owner, administrator, analyst and viewer roles; passkey sign-in.

  8. Proof of ownership

    A service is connected only after a DNS TXT record or a file on the site. Lockryn cannot be pointed at someone else's systems.

Security and privacy of the platform itself

Lockryn sees what it protects, which makes it a valuable target. The platform is built so that a fault in one part does not open up the rest.

How the platform is protected

  • Tenant isolation

    Every record, event and AI request belongs to one tenant. Isolation is enforced by database row-level security and a mandatory event-store filter, and tests check it for every endpoint.

  • Data in the EU

    The core, storage and backups live in data centres in Germany and Finland. Raw events do not leave the EU.

  • AI does not act directly

    AI has read-only tools. Actions are taken by a separate module after a policy check; anything unknown or incomplete is a denial.

  • Hash-chained audit log

    System and human actions are append-only. Changing any record breaks the chain, and verification shows exactly where.

  • Kill switch

    One action puts a tenant into “notify only” mode: Lockryn blocks nothing by itself and only tells people.

  • Medical and financial profiles

    For such services the agent does not collect request bodies, cookies or the Authorization header, and cloud AI is off by default.

Who it is for. For teams that run services on the internet but have no 24/7 security operations centre.

  • Web services and APIs

    Websites, APIs, mobile app back ends. Password brute force, scans and outages are noticed and contained, even when nobody is awake.

  • SaaS teams

    Several servers and environments, one console, roles for the team, a daily digest and a clear explanation of every incident.

  • Healthcare providers

    The “medical” profile: minimal data, cloud AI off, automatic actions only as recommendations a human approves.

  • Public institutions

    Data in the EU, an immutable audit log, a webhook to your own messenger instead of Telegram. A self-hosted licence is planned.

Questions and answers

Can I buy Lockryn already?
Not yet. The product is in development (MVP): we protect our own services with it first and are inviting a few companies to a pilot. There are no prices yet.
Can AI block something by itself?
No. AI has read-only tools: it explains the incident and proposes actions from the catalogue. A separate response module carries them out, and only if the policy allows. Critical actions need a human's approval.
Which actions does the first release take?
Only notifications and temporary IP blocks that lift themselves. Administrators, monitoring, payment gateways and search bots on the allow list are never blocked. New actions start as a dry run: the system records what it would have done.
Where is the data stored and who sees it?
In data centres in the EU (Germany, Finland). Cloud AI in an EU region receives de-identified summaries only; for medical, card and financial services it is off by default. The providers are named in the privacy policy.
What do I need to install?
Open-source agents go on your Linux servers — for logs, file integrity, attack detection and metrics; pilot participants get the exact list. Data travels to the core over encrypted, mutually authenticated channels, and the agents open no ports to the internet.
Can Lockryn be used against other people's systems?
No. A service is connected only after you prove ownership, and any action outside your infrastructure — “hack back”, scanning other addresses — is forbidden and not implemented.
Which languages does the console support?
The console and notifications are in Ukrainian and English. Polish, German and Spanish are planned.
Can I run Lockryn on my own servers?
A self-hosted licence is planned for later stages. Early access is an online service (SaaS) in the EU.

07:30morning review

Want to try Lockryn on your services?

We are taking on a few companies for the pilot. Tell us what you need to protect, and we will reply in person.

Get early access

Morning review · 07:30illustration

What happened overnight

  • 03:14web-1 · SSH password brute forceIP_1 temporarily blocked for 24 hours; nobody cancelled the block.
  • 03:15Telegram notificationTo the on-call admin, in their language, with no internal addresses.
  • 05:40api-1 · vulnerable packageA new vulnerability affects the installed version. Recommendation in the console.
  • 06:02shop.example · TLS certificateValid for another 61 days.

Waiting for a human decision: package update on api-1.