Privacy Policy

Last updated: April 11, 2026

Summary

AXME Code keeps all of your project data exclusively on your local machine — source code, decisions, memories, session transcripts, safety rules. We never upload project content to our servers.

Starting in version 0.2.7, AXME Code also sends a small amount of anonymous usage telemetry (install and startup events, product health counters, bounded error classes) to help us measure adoption and detect silent failures. Telemetry is fully opt-out with a single environment variable, and it never contains project content, file paths, hostnames, or any personally identifying information.

What data AXME Code creates locally

When you run axme-code setup, AXME Code creates a .axme-code/ directory in your project root containing:

  • Project knowledge base (oracle, decisions, memories)
  • Safety rules configuration
  • Session metadata and handoffs
  • Worklog of session events

All of this data lives exclusively on your local filesystem. It is gitignored by default and never leaves your machine.

Anonymous telemetry

AXME Code sends a small set of events to https://api.cloud.axme.ai/v1/telemetry/events. These events help us understand whether people are successfully installing, running, and getting value out of the tool.

What we send:

  • Lifecycle events — install (first run on a new machine), startup (every MCP server / CLI start, once per process), update (version changed since last run)
  • Product health events — setup_complete (did axme-code setup succeed or fall back to deterministic mode) and audit_complete (did the background session auditor produce extractions, how long it took, approximate LLM cost)
  • Error events — a bounded error category (audit, setup, hook, mcp_tool, auto_update) and a short slug from a fixed vocabulary (e.g. network_error, timeout, parse_error)

Each event carries these common fields: AXME Code version, source (binary or plugin), OS family (linux / darwin / win32), CPU architecture, a CI flag, a random machine ID (see below), and a timestamp. Health events add numeric counters (e.g. memories saved, decisions saved, audit duration in milliseconds, cost in USD, scanners run / failed).

What we never send:

  • Source code, transcripts, prompts, decisions, memories, safety rules, or any project content
  • File paths, directory names, project names, or the working directory
  • Hostnames, usernames, environment variables, or command-line arguments
  • Raw exception messages — errors are mapped to a small fixed set of slugs before sending, so they cannot leak paths or identifiers
  • IP addresses — the ingestion endpoint sees your IP at the network layer (any HTTP request does) but does not store it in the telemetry database
  • Anything that could identify you, your organization, or your project

Machine ID

On first run, AXME Code generates a random 64-character hex identifier (the machine ID) and stores it at ~/.local/share/axme-code/machine-id with file mode 0600. The ID is generated from crypto.randomBytes(32) and is not derived from any hardware identifier (no MAC address, no CPU serial, no hostname). It cannot be linked back to you or to any other service. If you delete the file, the next run will generate a new ID and we will count it as a separate install.

How to disable telemetry

Telemetry is entirely opt-out. Set either of these environment variables and AXME Code will send no events, generate no machine ID, and make no network requests from the telemetry module:

export AXME_TELEMETRY_DISABLED=1
# or the industry-standard:
export DO_NOT_TRACK=1

CI environments are detected automatically (CI, GITHUB_ACTIONS, etc.); events from CI are tagged with ci: true and excluded from our active-user counts.

LLM API usage

During axme-code setup, if you have a Claude API key or Claude subscription, AXME Code uses the Claude Agent SDK to scan your project and build the initial knowledge base. This runs through Anthropic's API under your own credentials and is subject to Anthropic's privacy policy. If no API access is available, AXME Code falls back to deterministic (no-LLM) detection.

The background session auditor also uses the Claude API (under your credentials) to extract memories and decisions from session transcripts. Transcripts never leave Anthropic's API — AXME Code does not forward them anywhere else.

Third-party services

The MCP server runs locally over stdio. The only external services AXME Code talks to are:

  • Anthropic API (under your credentials) — for LLM scanners and the session auditor, subject to Anthropic's privacy policy
  • GitHub Releases — for binary install and background auto-update checks, subject to GitHub's privacy statement
  • AXME telemetry endpoint (api.cloud.axme.ai) — only for the anonymous events described above, opt-out via environment variable

Data retention

Telemetry events are retained in our analytics database indefinitely in aggregate form. Because every event is anonymous and cannot be linked to a person, we have no concept of "your data" to delete on request. If you want to stop contributing future events, set AXME_TELEMETRY_DISABLED=1 — there is nothing else to do.

Open source

AXME Code is open source under the MIT license. You can audit the complete client-side telemetry implementation at src/telemetry.ts and the full technical specification at docs/TELEMETRY_TZ.md to verify every claim on this page.

Website (code.axme.ai)

This website is a static site hosted on Vercel. It does not use cookies, analytics, or tracking scripts. Vercel may collect standard web server logs (IP address, user agent) as part of its hosting infrastructure, subject to Vercel's privacy policy.

Contact

For questions about this privacy policy: contact@axme.ai