Artifactize

Privacy policy

Effective date:

Artifactize is an open-source command-line tool, licensed under Apache-2.0 and published by Hyukjin LIM (lhj6102). For privacy questions, contact the publisher through GitHub issues. Do not post secrets in public issues.

The publisher collects nothing

Artifactize has no telemetry, analytics, automatic crash reporting, publisher accounts or automatic update checks. No server operated by the publisher receives data from the tool.

Local data

Run history, review results, reuse caches, captured output and saved Agent conversations stay in your local state directory unless you choose to share them. Codex sign-in tokens and shared review server credentials are stored there too.

--state-dir or ARTIFACTIZE_STATE_HOME overrides the location; XDG_STATE_HOME also takes precedence on Windows. OpenAI and Anthropic API keys are read from OPENAI_API_KEY and ANTHROPIC_API_KEY, not saved by Artifactize. An optional ARTIFACTIZE_CODEX_AUTH_FILE points to an existing sign-in file that Artifactize only reads.

When the desktop app cannot find an existing artifactize command in standard terminal locations or its inherited PATH, it sets up the bundled commands locally. On macOS it requests administrator approval to write the app's command directory to /etc/paths.d/artifactize; a cancelled prompt is remembered for that bundle location in desktop-cli-declined.json in the state directory. On Linux, an AppImage copies the commands and a version marker to $XDG_DATA_HOME/artifactize/bin (otherwise ~/.local/share/artifactize/bin) and creates links in ~/.local/bin, without administrator rights. It does not edit shell startup files or replace unrelated command links. These steps send no data to the publisher.

Providers you configure

When you run an Agent eval, Artifactize sends its instructions, declared Artifact context and the project content returned through its declared read-only tools to the configured provider: OpenAI, Anthropic or ChatGPT/Codex through your sign-in. Agent follow-up messages and conversation context go to the same provider. Model-listing commands contact the selected provider; Codex sign-in, token refresh and sign-out contact OpenAI.

That provider's own privacy policy governs the data it receives: OpenAI or Anthropic.

Runtime and Human evals do not call LLM providers. User-declared commands and tools are trusted local programs, not a network sandbox: they can communicate externally, and Human tools can open web URLs.

Your shared review server

If you run or connect to artifactize server, verdicts and review metadata are shared with that server, operated by you or your team. The default summary includes owner fields, fingerprints, declared profiles, reviewer/producer names and timestamps; choosing full sharing also sends captured output. See shared review server details.

Removing data

Delete the state directory printed by artifactize doctor to remove local history, caches and stored sign-ins. API keys in your environment and an external Codex auth file are separate. Removing an unpackaged install keeps its state directory. Uninstalling the Store package removes its package-private state. Project files, exported files, files written by user-configured commands outside the package-private view, existing unvirtualized AppData files and state explicitly placed outside AppData are not removed with the package; delete those separately if no longer needed. See cleanup and uninstall.

Changes

This policy is versioned in the public repository. Changes will update the effective date above.