Naevik is an AI-agent endpoint security product operated by SensiSec Pty Ltd (ABN 88 663 013 063), a company registered in Australia (the “Company,” “we,” “us”). Our corporate site is sensisec.com; the Naevik product site is naevik.ai. For privacy questions, contact privacy@naevik.ai.
Naevik watches what AI agents do on your machine, not what you do. The agent enforces policy on what AI coding tools and desktop assistants (Claude Code, Claude Desktop, Cursor, GitHub Copilot, OpenClaw, and similar) are allowed to run, read, write, and connect to. It does not read your file contents, does not inspect your network payloads, and does not capture keystrokes, screen contents, or browsing history.
Your endpoint data stays on your device. Audit records — the log of what agents run, read, write, and connect to — live in a local database on the machine. In Personal, there is no server and that data never leaves the device. In Team, your organization runs its own Naevik server and only the data it configures is sent there — not to us. Only two kinds of agent data can reach SensiSec: opt-in crash reports (macOS only, off by default; see Section 5) and, in Personal, an install count — a random, pseudonymous install id sent with the daily update check (on by default, and you can turn it off; see Section 6). Neither contains any endpoint event data. A small amount of other data — a purchase or support email — may reach us too; those are described in the sections below.
Naevik ships in two editions from the same agent:
This policy describes two things:
In Personal, you are both the user and the operator: for the endpoint event data on your device there is no third-party controller. (We remain the controller for the limited data we do receive — a purchase, support, or crash report, the install count, and website requests — described below.) In Team, your organization is the data controller for endpoint event data and SensiSec is a data processor acting under their instructions. The website we operate ourselves as controller.
Naevik Agent uses the operating system's built-in enforcement hooks to evaluate actions taken by processes on the device against the configured policy, before those actions take effect.
Events evaluated may include:
The agent does not read file contents as part of policy enforcement. Network decisions are made on flow metadata only. The agent does not decrypt TLS, does not read application-layer payloads, does not perform deep packet inspection, does not capture keystrokes, screen contents, or browsing history, and does not alter DNS responses. Network connections initiated by AI agents are evaluated against policy; your ordinary web browsing traffic is passed unmodified and is not subscribed to.
Process arguments and file paths may contain sensitive values depending on how a tool is invoked or how a project is structured. You (Personal) or your organization (Team) control what metadata is logged, retained, redacted, or — in Team — uploaded.
Audit records are written to a local database on the device.
Local-only events. When the agent is configured to allow developer-authored local policy overlays, events generated by those local rules are stamped non-uploadable at the moment of capture and remain on the device. Toggling any upload setting later does not retroactively reclassify previously captured records — the consent boundary is fixed at capture time.
When your organization uses Naevik Team, the Naevik agent on each enrolled computer sends your organization’s Naevik server the computer’s name, operating system, IP addresses and agent version, the security events Naevik records, the computer’s serial number and model, and when it last started.
Unless your organization turns it off, it also sends the name of the account signed in at the computer’s screen (on Windows with its domain and, when Windows provides it, the account’s sign-in name, usually a work email address). The server keeps only the latest of these values, not a history of who used the computer. When your organization turns the account name off, the server deletes it for computers that are online within seconds and for others when they next connect; a computer that never connects again keeps its last account name until your organization removes it from Naevik.
Your organization’s identity provider may also send the server its staff list — for each person their name, work email, user name, the identifier the identity provider uses for them, and whether they are active — so administrators can see whose computer each one is and who does not have Naevik yet. People on that list cannot sign in to the Naevik console; only people an administrator invites or gives access can.
All of this goes only to your organization’s Naevik server, which your organization runs; SensiSec does not receive it. Naevik Personal sends none of it.
Unless your organization turns it off, the Naevik agent also checks how the AI coding tools on the computer (Claude Code, Codex and Cursor) are configured, and sends your organization’s Naevik server, for each account on the computer that has their settings: the account’s name, the folders of the projects those tools list, which risky settings are on (for example, a tool set to skip every permission prompt) and, where available, the tools’ versions; and the MCP servers the tools are set to start — for each, its configured name, how it connects, what it runs (a package name, a program’s file name without its folder, or a web address reduced to its scheme, host and port), whether it is turned off, and whether a credential is written into its configuration.
The agent never sends the contents of these files, the values of the settings, or any credential, header, environment variable or command-line argument from them. All of this goes only to your organization’s Naevik server, which your organization runs; SensiSec does not receive it. Naevik Personal does not run this check. This check has its own setting: turning off the signed-in account name described above does not turn it off.
SensiSec does not operate a multi-tenant cloud that aggregates your endpoint event data. We do not receive your process, file, or network event records from the agent, we do not have access to a Team customer’s server, and we do not sell, share, or use endpoint data for any purpose. The only exceptions are the opt-in crash reporting and the install count described next, and neither contains endpoint event data.
Naevik can send macOS-generated crash reports for the agent to help us find and fix defects. This is off by default and never runs unless you turn it on.
In Personal, the agent makes one outbound HTTPS request to naevik.ai when it starts and every 24 hours, to fetch update information and security, end-of-life, and feature notices. The request carries the Naevik version and platform in its User-Agent (for example Naevik/1.0.25 (windows)). You can turn this check off with Check for updates and security notices: on macOS in Settings → General, and on Windows in Settings → About → Updates (Windows asks for administrator approval). Linux has no Settings app; there, use the configuration file described in Section 6.2. When the check is off, no request is made at all.
When an update is available, the installer is downloaded from naevik.ai/dl. That download does not carry the install id described below.
When you click Get latest rules in the rule library, Naevik downloads the list of built-in rule groups from naevik.ai. This request is made only when you click. Like the update check, it carries only the Naevik version and platform in its User-Agent. It does not carry the install id, and it sends nothing about your rules, your device, or your activity.
So that we can count how many installations of Naevik are in use, and on which platforms and versions, the update check also carries an install id in a request header.
In Team, when the agent installs a software update, it reports update success or failure along with version identifiers to your organization’s server. This is a small JSON record (status, reason, timestamp, version) used to support staged rollouts and troubleshooting. It contains no endpoint event data and no file, process, or network activity metadata. In Personal, the update result is not sent to us or to anyone else.
The website is a static page hosted on a content-delivery network. We do not load third-party analytics, advertising trackers, social-media pixels, A/B-test scripts, or session-replay tools.
Naevik Personal is now free, and we no longer sell it. While it was sold, checkout and payment were handled by a third-party payment provider acting as merchant of record; your payment-card details were processed by them, not by us, and we never received or stored your full card details. We received your email address and order metadata so that Naevik could email you your license key and provide support — the license email comes directly from us.
If you bought a license, we store your purchase email encrypted at rest, linked to your license. We use it only to:
We do not send newsletters or marketing email, and we do not sell or share your address. Outside of license delivery, support replies, and urgent security notices, we aim to email you no more than about twice a year. If we ever add an optional product newsletter, it will be separate and strictly opt-in.
On the device, the agent retains audit records in the local database subject to two bounds: a configurable time window (on the order of weeks by default) and a row cap of 100,000 records. Whichever bound is reached first triggers a sweep of the oldest records. Records acknowledged as uploaded (Team) and non-uploadable local-only records are retained under the same window and swept by the same paths. You can shorten the window in the menu app, or delete all local event history at any time from Settings → Privacy on macOS or Settings → Event history on Windows.
For crash reporting, a small on-device ledger records the identifiers of reports already sent so the same crash is not uploaded twice.
Install-count records (Section 6.2) are kept for about three months. After that we keep only aggregate counts, such as the number of active installations per platform and month, with no install ids.
In Team, server-side retention is configured by your organization on its own server; SensiSec does not operate that server and does not set its retention. Any email address you give us (purchase, support, or waitlist) is stored encrypted at rest and retained until it is no longer needed for that purpose or until you ask us to remove it, whichever is sooner.
The agent ships as a code-signed, notarized binary on macOS and as a signed package on Linux and Windows. Updates are verified against publisher identity, signature, and content hash before installation. On macOS, both the enforcement and network-filtering components run as system extensions installed under Apple’s standard user-approval flow; no system approval prompts are bypassed. Communication with a Team server uses an encrypted (HTTPS) connection with a per-device token.
In Personal, your endpoint data stays on your own device under your control; you can inspect or delete it from the menu app at any time. For data we hold (a purchase, support, or waitlist email, an opt-in crash report, or install-count records), or for general questions about Naevik’s data practices, contact privacy@naevik.ai and we will respond within thirty days.
Install-count records are tied only to the random install id, not to your name, email, or IP address, so we cannot tell which records are yours. The practical way to stop new records being linked to your earlier ones is to delete the id file (Section 6.2). Turning the install count off stops the id being sent at all.
In Team, your organization is the data controller for endpoint event data and is the right party to contact for access, correction, deletion, or objection requests under applicable law (GDPR, the Australian Privacy Act 1988, and similar regimes).
Naevik is a developer tool not directed at children. The website does not knowingly collect information from anyone under 16.
SensiSec is based in Australia. The website CDN and email infrastructure may process data outside Australia under their own terms. Endpoint event data does not flow to SensiSec; in Team its location is determined by where your organization hosts its server, and in Personal it stays on your device.
We may update this policy as the product evolves. The “Last updated” date at the top reflects the most recent change. Material changes will be announced on this page and, where appropriate, by email.
SensiSec Pty Ltd (ABN 88 663 013 063)
Australia
privacy@naevik.ai