[velixio]

Replace the Python application runtime with native Go API, worker, scheduler, MCP and migration binaries. Python remains in the internal browser service for Playwright and browser-dependent scraping.

The port covers authentication and workspace isolation, settings and encrypted secrets, workbooks and enrichment, leads and signals, audiences and destinations, automations, outreach, governance and agent access. PostgreSQL with forced workspace RLS owns data and the durable queue; Redis carries progress. The Go migration runner adopts the exact Alembic baseline e7b2c4d6f8a0 and applies migrations through version 4.

Final backend fixes

• Scope person-dependent provider caches to their inputs and settings; do not reuse shared-domain lead scores.
• Recognize known free native providers while preserving explicitly declared prices.
• Save multi-field encrypted provider settings atomically, with rollback and encryption-failure regressions.
• Write destination insert timestamps in UTC, covered under UTC, Asia/Kolkata and America/New_York.
• Build filesystem statistics helpers on Darwin and Linux.

README and deployment notes

The README now includes light/dark dashboard screenshots and two compressed GIFs showing connected records and an import review. These are explicitly labeled as previews from the separate dashboard redesign; this PR does not bundle that local redesign or canonical-record migration 5.

The server example uses a reviewed Go revision instead of assuming an older release tag contains this runtime. Existing Python installations have a separate upgrade section linked to the cutover runbook.

Release images use explicit opengtm-{api,worker,mcp,migrate,scheduler,browser} names. The plain legacy ghcr.io/debpalash/opengtm alias stays unchanged.

Validation

• Full CI passed on backend fix commit f6abd818: Go vet, race unit/integration tests with PostgreSQL and Redis, shipping builds, frontend lint/tests/build, and browser-service checks. The integration suite includes the real API/worker scenarios with no Python application runtime.
• All final-head checks passed on 0d973725: Go vet, race unit/integration tests and shipping builds; frontend lint/tests/build; browser-service lint/tests. No backend code changed in the README/assets commit.
• Targeted local race regressions and generated-SQL consistency checks passed. GIFs were captured from the rendered local app using fictional records; both are under 3 MiB, and README asset links resolve.
• Local native cutover/restore rehearsal passed against the versioned e7b2c4d6f8a0 reference schema using tracked Python source b8bfc57: Python login and workbook creation, verified PostgreSQL/data-directory backups, Go adoption and job tenant/executor backfill, legacy-token authentication in Go, native workbook jobs, then backup restoration and Python login/workbook access. The restored queue returned to its pre-cutover Python state. Fixture restore plus Python restart took 1.83s; this is not a production recovery-time estimate.

Merge checks

• Publish isolated backend fixes and README assets.
• Document explicit release image naming.
• Pass the isolated native reference-schema cutover/restore rehearsal.
• Confirm all checks on the final head.

Production cutover preflight

• Perform the deployed Docker staging cutover/rollback drill with the installation's actual previous Python release and data. The local reference-schema fixture does not verify upgrading an older Alembic schema or running the full container topology.

Existing Python production cutovers must retain a verified pre-cutover backup and pass the rollback rehearsal before deployment. See migration ownership and preconditions and Python-free cutover. No release tag, image publication or production deployment is part of this PR merge.