Your codestays yours.

Registry requests: When you run axint search or axint add, the CLI makes an HTTPS request to registry.axint.ai with the query or package name you provided. Standard request logs (timestamp, IP, user-agent) are retained for abuse prevention and aggregated into anonymous download counts.

Publishing: If you publish a package, we store your GitHub ID, username, and avatar. Package metadata and source are public by design.

axint Cloud: If you opt in, Cloud stores the reports, package references, and workspace metadata you explicitly create. Signed-in collaboration features also store your GitHub identity. You can delete your workspace and associated data at any time.

Website: We use Vercel Analytics to track anonymous pageviews. No cookies, no user IDs, GDPR compliant.

Email capture: If you choose to leave an email address for Axint updates or Cloud early access, we store that submission in our lead inbox and, when configured, in Axint-owned KV storage only so we can reply about Axint. Analytics events record that the form worked, not the submitted email address.

• No source-code telemetry from axint compile — compilation still happens locally
• No source-code telemetry from the CLI — automatic feedback packets are source-free diagnostic summaries
• No prompts, generated Swift bodies, command arguments, local file paths, or file names in CLI/MCP adoption telemetry
• No raw Registry searches, project descriptions, or project names in project-interest analytics
• No device fingerprinting, machine IDs, or hardware identifiers
• No email address required to install or use the compiler
• The CLI transmits no source code — ever — unless you explicitly publish a package or opt in to a source-sharing Cloud workflow. The axint.ai playground is the one exception: it compiles server-side (see below)
• No ads, tracking pixels, or third-party data sharing

axint compile:The compile step runs 100% locally. TypeScript in, Swift out, in-process. Your source code doesn't leave your machine. The CLI may send a separate source-free adoption heartbeat so we can see whether terminal and agent installs are working. That heartbeat contains command class and version only, never source, prompts, generated Swift, arguments, file names, or local paths.

Adoption telemetry: Axint CLI and MCP can send a tiny usage heartbeat with Axint version, command class, MCP tool name, coarse host hint, OS family, Node major version, CI flag, and a random anonymous install ID. Search and project-brief signals are reduced on the client or at the Registry boundary to allowlisted categories, goals, feature areas, lifecycle stage, delivery target, complexity, Apple capability tags, query-length buckets, and result buckets. Raw searches, descriptions, project names, source, and paths are not stored. Inspect telemetry with axint telemetry status, choose product signals only with axint telemetry standard, enable source-free diagnostic learning with axint telemetry enhanced, turn it off with axint telemetry opt-out, or disable it per process with AXINT_TELEMETRY=off.

Source-free feedback: When Cloud Check, axint Run, Repair, or Prove finds an Axint learning signal, the CLI queues an anonymized packet locally. Enhanced sharing can submit diagnostic codes, failure class, compiler version, project shape, repair status, redacted and truncated issue text, expected/actual behavior, failure excerpts, deterministic-repair outcome, and build/test/runtime proof. It does not include raw source, generated Swift bodies, credentials, or local file contents. Axint-owned projects may carry an internal dogfood marker so Pulse can separate them from external reports. Opt out with axint feedback opt-out or AXINT_FEEDBACK=off.

axint search / add / publish: These commands talk to registry.axint.ai because that's their job — discovering, installing, and sharing public packages. They run only when you invoke them, and send only what's required for that specific command. Registry search demand is retained only as coarse themes and result-count buckets; the submitted search text is not retained in product analytics.

axint Cloud: Cloud validation is opt-in. Nothing is sent unless you choose to run a Cloud report or save a workspace-backed artifact. Signed-in collaboration features add GitHub identity to that flow. If you never touch the Cloud features, no data ever leaves your machine.

MCP Server:The local MCP server bridges your IDE to the local compiler. Compilation stays on your machine; the server itself doesn't upload your source.

The playground at axint.ai/#playground is the one part of the site that runs the compiler on our servers instead of in your browser. This is an architectural trade-off: shipping the TypeScript compiler to the browser would add roughly 10MB to the page weight, so we compile server-side instead. The behaviour is identical to the CLI — same IR, same Swift output.

How it works: As you type, the editor sends your source to a Next.js server action over HTTPS. The action calls the same compiler the CLI uses, in-process, and returns the generated Swift. Source is capped at 200KB per request.

What we don't do: Source is held in memory for the duration of the compile and discarded — never written to a database, never written to persistent logs, never forwarded to a third party. There is no analytics on what you type. Request logs at the Vercel edge capture the usual timestamp, IP, and user-agent for abuse prevention, but not the request body.

If you'd rather not: Install the compiler with npm install -g @axint/compiler and run it locally. The CLI and the playground produce identical output.

Published Templates: The registry stores metadata and source code for templates you choose to publish. This data is intentionally public so others can discover and use your templates.

GitHub OAuth: We use GitHub OAuth to authenticate registry publishers. We store your GitHub ID, username, and avatar URL. We do not harvest or store your email address.

Opt-In Publishing: Nothing is published to the registry unless you explicitly choose to publish it. Your local work is never synced or shared.

Vercel Analytics: This website (axint.ai) uses Vercel Analytics to track anonymous pageviews, referrers, popular pages, and high-level product events such as GitHub star clicks, email-capture submissions, partner-form submissions, and Cloud save actions. No personally identifiable information is collected in those analytics events.

Axint first-party analytics: We also send source-free product events to Axint-owned infrastructure so we can understand which docs, Cloud actions, and Registry flows are working. These events may include page path, referrer, coarse country, browser user agent, anonymous browser/session IDs stored in localStorage, and high-level action names. They do not include code, prompts, generated Swift, or private project data.

No Cookies:We do not set tracking cookies. Vercel Analytics respects “Do Not Track” headers and is GDPR compliant. The Axint first-party event bridge also respects Do Not Track and Global Privacy Control signals, and uses localStorage identifiers instead of cookies.

axint is licensed under Apache 2.0, which means the full source code is public and auditable. You can read the compiler and CLI end-to-end, confirm that axint compile makes no network calls, see exactly what axint search, add, and publish send to the registry, and build the whole toolchain yourself.

Code + transparency: The source is available at github.com/agenticempire/axint. If you want to verify a privacy claim, the code is the source.

Registry Metadata: Published package metadata is retained indefinitely to preserve the registry. If you delete a package, the metadata is removed.

Analytics: Vercel Analytics data is retained for 30 days by default.

Product signals: Anonymous adoption and project-interest events are retained for 90 days. Source-free enhanced diagnostic and proof packets are retained for 180 days. These limits are enforced by a nightly deletion job; neither dataset contains source code, project names, or local paths.

Compiler: No data is retained by the compiler — it generates output and exits.

Access:You can request a list of packages you've published and associated metadata.

Deletion: You can delete your published packages and request removal of associated account data.

Opt-Out:You can opt out of website analytics by enabling “Do Not Track” in your browser.

To exercise these rights, contact privacy@axint.ai.