Skip to content
Home › Get Started › Quick Start

Quick start

This is the five-minute happy path. By the end you'll have an AMX install configured against one database and one LLM, with a generated description ready to write back to the DB.

1. Install

pip install amx-cli

2. Open the interactive session

amx

You'll land in the AMX REPL. Slash commands drive everything from here.

3. Configure profiles

Run the first-time setup wizard:

/setup

The wizard walks you through:

  1. Database connection. Pick an engine (PostgreSQL, Snowflake, Databricks, BigQuery, …), then enter that engine's connection details. The profile is saved to ~/.amx/config.yml.
  2. LLM provider. Pick a provider (OpenAI, Anthropic, Gemini, …) and supply the API key. Keys are stored in the OS keychain when available.
  3. Optional data sources. Document roots and codebase paths for the RAG and Code agents. You can skip these and add them later with /add-doc-profile and /add-code-profile.

Confirm everything is working:

/config
/db
/connect

/connect performs a real connection test against the active DB profile.

4. Run the agents

/run

/run opens a scope picker — Database / Schema / Asset — then dispatches the active agents (Profile, RAG, Code) and merges their suggestions. Mid-flight you'll see live progress with per-column status.

For a specific table:

/run sap_s6p.t001

For multiple tables:

/run t001 vbak

5. Review and apply

When the agents finish, /run opens its interactive review. First you choose a review strategy — one-by-one, accept-all-high, accept-all, or reject-all. For each column you see:

  • The top suggestion with confidence band (high / medium / low) and logprob score.
  • A few alternatives.
  • The evidence sources behind each suggestion (DB profile, code references, doc snippets).

When reviewing one-by-one, AMX presents a numbered picker: type a number to pick a description, s to skip, or o to write your own text.

When you're done reviewing, write back to the database:

/apply

6. Inspect what happened

/history list
/history show <run_id>
/usage 7d

/history list shows recent runs with duration and model. /history show dumps the full JSON payload (scope, metrics, tokens, results, errors). /usage summarises token usage over a window.

When something goes wrong

Open AMX and run /doctor:

amx

> /doctor

/doctor diagnoses install / config / connectivity and prints actionable hints next to each ✗.

For the full review experience and a real example, continue with the first run walkthrough.