Authentication and Audit

Authentication & Audit

golgi is multi-user and keeps a tamper-evident audit trail of every meaningful action. Auth and audit are per-user-local (a single workstation/server), not a multi-tenant cloud service.

Source: golgi/auth/ — models.py (DB), session.py (session), decorators.py (gating + logging), audit.py (the audit writer).

---

Users & sessions

Users live in a local SQLite database (auth.db), with email, optional username, hashed password, profile fields (name, country, institution, position), and an avatar. The GUI's Sign in dialog authenticates against this DB; the session is held process-side. Projects carry an owner and a shared-with list, so the welcome screen shows you only your own and shared projects.

In headless mode (the Python API / CLI) golgi synthesises a local headless user so the auth-gated pipeline drivers run without interactive login — the audit writer still records the run.

Gating & logging decorators

Pipeline actions are wrapped with two decorators:

• @log_action("name") — emits an audit row (with truncated arg capture) recording success or failure.
• @gated("name") — refuses to run when no user is logged in, surfaces the login prompt, and audits the blocked attempt.

Audited actions include load_geometry, build_mesh, solve_fem, solve_fiber_paths, sweep, export_study, import_study, segment_uct, export_figure, bulk_export, and generate_report.

The audit log

audit.py is a flight recorder: actions push events onto a bounded in-memory queue that a daemon thread batches into the database. It's robust by design — if the DB write fails, events fall back to an append-only audit_fallback.jsonl, and that file is drained back into the DB on the next successful write. Each event records timestamp, user, action, a JSON payload, the project directory, and a status (success / failure / info / blocked).

You can review the trail in the Activity tab of the Project Details dialog (GUI). When a study is exported, a project-scoped audit excerpt travels in the bundle; on import, those events are replayed under the importing user, tagged with the original provenance — so a shared study carries its history with it.

---

See also

Reproducible Study Bundles · GUI Walkthrough · Headless / HPC · Architecture