AMX — Agentic Metadata Extractor

AMX (Agentic Metadata Extractor) provides AI-powered guidance and reference for data analysts, data engineers, and data catalog owners working with undocumented database schemas.

AMX walks your database, reads your documentation and codebase, then drafts a complete description for every table and column, with confidence scores and a human review before anything lands in the live database.

Tip

AMX supports 10 database backends and 7 LLM providers. The fastest way to evaluate is pip install amx-cli, then run amx and walk through the /setup wizard. Five minutes from install to your first reviewed description.

Try AMX

• Install AMX

  Install from PyPI with the database backends you need. Optional extras keep the default install lean.

• 5-minute quickstart

  Install, configure, run agents, review, apply. The full happy path in one short walkthrough.

• First-run walkthrough

  Narrated end-to-end session against a sample SAP schema. Useful before running against production data.

• Run diagnostics

  amx doctor checks install, config, and connectivity from any shell — even when AMX itself can't start.

Explore AMX

• Architecture

  The CLI shell, multi-agent orchestrator, search agent, and storage layer.

• The three agents

  What the Profile, RAG, and Code agents read and how the orchestrator merges them.

• Universal Metadata Interface

  The backend-neutral entity model that lets ten databases share a single workflow.

• Human-in-the-loop review

  The review wizard, bulk-accept, write-back, and the full audit trail.

CLI reference

• Slash command map

  Every namespace and command, grouped by purpose.

• /run and /apply

  Run agents, review suggestions, write back to the database.

• /ask and /search

  Conversational metadata Q&A with grounded retrieval and live verification.

• /history

  Run history, comparison across runs, re-evaluation.

• amx doctor

  Install / config / connectivity diagnostics.

• Common flags

  --db-profile, --llm-profile, --apply, --csv, and the rest.

Backends

• All backends

  Capability matrix and per-backend deep dives.

• PostgreSQL

  Reference adapter. Standard COMMENT ON … write-back.

• Snowflake

  Account, warehouse, role; profiling-mode tuning per workload.

• Databricks

  Unity Catalog SQL warehouse, with TLS recovery for corporate networks.

• BigQuery

  Project + dataset, byte-scan controls for large tables.

• MySQL · Oracle · SQL Server · Redshift · ClickHouse · DuckDB

  Six additional backends with first-class object-type listings.

Operate AMX

• Configuration

  ~/.amx/config.yml schema, env vars, TLS and proxies, profiling modes.

• LLM providers

  Setup notes for each supported provider, including Batch mode.

• Data sources

  Codebase scans, document RAG, and the search catalog.

• Collaboration

  Shared history store, team setup, safety guards.

• Troubleshooting

  FAQ, common errors, and diagnostic recipes.

• Python API

  Stable amx.core surface for headless use from scripts and notebooks.

Quick links

• Changelog

  Release notes for every published version.

• Contributing

  Development setup, branching, commit format, release process.

• About the maintainer

  Omer Yasir Kucuk — the engineer behind AMX.

• Security

  How to report vulnerabilities. Supported versions and scope.

• GitHub repository

  Source code, issue tracker, releases.