Multi-Phase Pipeline
Every conversion runs through five phases: discovery, planning, conversion, validation, and auto-fix. Objects are converted in dependency order (views before the procedures that use them, utilities before their callers) so cross-file signatures are always real, never guessed, and every generated file gets a confidence/completion grade.
Platform Auto-Detection
Discovery classifies the source SQL dialect and runs a separate, deterministic, no-LLM regex pass across 133 markers to fingerprint the specific source platform (Databricks, Snowflake, PostgreSQL, MySQL, SQL Server, Oracle, Redshift, BigQuery, Teradata, or Hive) with a fully explainable confidence score.
Bulk Conversion & Resumable Sessions
Convert an entire directory of .sql files in parallel with configurable worker counts and scheduling strategy. Every run is tracked as a session, so interrupted or long-running bulk jobs resume exactly where they left off with `3xcode session resume` or `3xcode pyspark convert --resume`.
Web UI
A local, browser-based workspace launched with `3xcode ui`: upload and convert files, manage bulk sessions with live progress, browse and download output, and configure providers, all from a single-user localhost dashboard.
Multi-Provider & BYOK
Route conversions through Anthropic, AWS Bedrock, Google Vertex AI, or Azure AI Foundry. Bring your own provider key, stored locally via your OS keychain. Your source SQL is never sent to the 3XDE backend, only to the LLM provider you choose.
Organizations & Team Licensing
Create or join an organization to share visibility and, optionally, a shared license pool across teammates. Licensing resolves org-first with individual-fallback, and organization-shared licenses are exempt from the per-machine cap.