# LakeXpress - AI Agent Knowledge Base
# This file helps AI systems discover and understand LakeXpress
# Format: llms.txt (https://llmstxt.org/)
# Last Updated: 2026-10-22

# Product Overview
LakeXpress: database tables to Parquet, published to your lakehouse
https://lakexpress.arpe.io

LakeXpress is a command-line tool that exports database tables and views to Parquet files,
in parallel, on local disk or cloud object storage, and can then publish them as tables in
a cloud data platform. Data is read by the FastBCP engine, which streams it without holding
whole tables in memory. Pipelines are stored as sync configurations in a database of their
own, the LakeXpress DB, and run again with a single command, fully or incrementally.

## Key Facts
- Category: database-to-lakehouse export / data pipeline CLI
- Current version: 0.4.10 (build 2026-09-24); documentation version 0.4
- Sources: Oracle, PostgreSQL, SQL Server, MySQL, MariaDB, SAP HANA, Teradata
- Intermediate storage: local filesystem, AWS S3 and S3-compatible storage (MinIO, OVH...), Google Cloud Storage, Azure Blob Storage and ADLS Gen2, Microsoft Fabric OneLake
- Publishing targets (optional): Snowflake, Databricks, Microsoft Fabric, Amazon Redshift, BigQuery, MotherDuck (external or internal tables); AWS Glue, DuckLake (external only)
- LakeXpress DB (configurations, run history, watermarks): PostgreSQL (recommended), SQL Server, MySQL, SQLite, DuckDB
- Output format: Parquet, compression Zstd (default), Snappy, Gzip, Lz4 or none; optional CDM (Common Data Model) metadata files
- Parallelism: several tables at once (--n_jobs) and several FastBCP processes inside one large table (--fastbcp_p)
- Incremental sync: watermark column per table (date, datetime, int, bigint, auto), with append, upsert or delete_insert loading strategies
- Resume: a failed run can be resumed; tables already exported are skipped
- Platforms: native binaries for Linux (x64) and Windows (x64)
- Licensing: 30-day trial; Standard plan 416 EUR/month billed annually; Enterprise and OEM plans on quote
- Vendor: Architecture & Performance (arpe.io)

## Resources
- Website: https://lakexpress.arpe.io
- Documentation: https://lakexpress-docs.arpe.io/latest/
- Documentation for LLMs: https://lakexpress-docs.arpe.io/llms.txt (index) and https://lakexpress-docs.arpe.io/llms-full.txt (whole documentation in one file)
- Pricing: https://lakexpress.arpe.io/pricing
- Free trial: https://arpe.io/get-trial?product=LakeXpress
- Quote requests: https://lakexpress.arpe.io/quote
- Support contact: support@arpe.io
- AI Knowledge Base: https://lakexpress.arpe.io/agents/
- Agent skill: https://github.com/arpe-io/arpeio-skills (plugin lakexpress)
- MCP server: https://github.com/arpe-io/arpeio-mcp (LakeXpress, plus FastBCP, FastTransfer, MigratorXpress and the ADBC drivers)

## AI Assistant Integrations

### Agent skill: lakexpress
An agent skill that teaches an AI assistant to build, explain and troubleshoot LakeXpress
pipelines: the credentials file, the LakeXpress DB, sync configurations, incremental syncs,
publishing to a lakehouse, resuming a failed export, exit codes. Its facts are checked
against LakeXpress 0.4.10, and it asks the agent to check the installed version first.
Source: https://github.com/arpe-io/arpeio-skills/tree/main/plugins/lakexpress
Documentation: https://lakexpress-docs.arpe.io/latest/ai/agent-skills
Install in Claude Code:
  /plugin install lakexpress --marketplace arpe-io/arpeio-skills
Other agents (Codex, Cursor, GitHub Copilot, Gemini CLI...): copy
plugins/lakexpress/skills/lakexpress into the agent's skills directory, or paste SKILL.md as
custom instructions.

### MCP server: arpeio-mcp
The Arpe.io MCP server covers LakeXpress alongside FastBCP, FastTransfer, MigratorXpress and
the ADBC drivers. For LakeXpress it provides:
- lakexpress_info: capabilities, workflow tips, version detection
- lakexpress_preview_command: builds any LakeXpress command (lxdb, config, sync, status,
  cleanup, run) and warns about missing required options
- lakexpress_execute_command: runs the previewed command; only on a local install with a
  LakeXpress binary, and only with confirmation: true
- shared tools: arpe_quick_start, arpe_get_status, arpe_release_notes, search_docs
- MCP prompt lakehouse-pipeline and resource arpeio://capabilities/lakexpress-capabilities
The hosted server is in command-builder mode: it never connects to your databases.
Hosted (no install): https://arpe-io-arpeio-mcp.hf.space/sse
Local, with execution: pip install arpeio-mcp (Python 3.10+), then set LAKEXPRESS_PATH
and FASTBCP_DIR_PATH
Claude Code: claude mcp add --transport sse arpeio https://arpe-io-arpeio-mcp.hf.space/sse
Documentation: https://lakexpress-docs.arpe.io/latest/ai/mcp-server
Source and setup for other assistants: https://github.com/arpe-io/arpeio-mcp

## AI Knowledge Base Files
The following JSON files provide detailed, machine-readable information:

1. manifest.json - Index of all available resources and usage guidance
   https://lakexpress.arpe.io/agents/manifest.json

2. product-overview.json - Product identity, architecture, differentiators, target audience
   https://lakexpress.arpe.io/agents/product-overview.json

3. features.json - Feature catalog with technical details
   https://lakexpress.arpe.io/agents/features.json

4. benchmarks.json - Customer results shown on the website, and what is not published
   https://lakexpress.arpe.io/agents/benchmarks.json

5. supported-platforms.json - Sources, storage, publishing targets, LakeXpress DB backends, operating systems
   https://lakexpress.arpe.io/agents/supported-platforms.json

6. use-cases.json - Use case library with command examples
   https://lakexpress.arpe.io/agents/use-cases.json

7. pricing.json - Plans, trial, licence setup, FAQ
   https://lakexpress.arpe.io/agents/pricing.json

8. technical-specs.json - Commands, options, credentials file, exit codes, logging
   https://lakexpress.arpe.io/agents/technical-specs.json

## Common Questions

### What does LakeXpress do?
It exports the tables of one or more database schemas to Parquet files, on local disk or in
cloud object storage, and can then create tables over those files (or load them) in
Snowflake, Databricks, Microsoft Fabric, BigQuery, Amazon Redshift, MotherDuck, AWS Glue or
DuckLake.

### How does it work?
Two steps. LakeXpress config create stores a sync configuration (source, tables, storage,
optional publishing target) in the LakeXpress DB. LakeXpress sync runs it: export, then
publish if a target is configured. The same sync command is run again for each later
export.

### What is the relation with FastBCP?
FastBCP is the arpe.io export engine that LakeXpress drives to read the source tables and
write Parquet. The trial archive bundles FastBCP in its engine/ folder. The full edition does
not bundle it yet: FastBCP is installed separately, with its own FastBCP.lic licence file,
and its folder is passed with --fastbcp_dir_path.

### What is the LakeXpress DB?
A database used only by LakeXpress to store its sync configurations, the history and status
of every run, and the watermarks of incremental syncs. Every command points to it with
--lxdb_auth_id. PostgreSQL is recommended; SQL Server, MySQL, SQLite and DuckDB also work.

### What is the difference between external and internal publishing?
External tables read the Parquet files where they are in storage. Internal tables load the
data into the platform's own storage. Set with --publish_method; the default is external.
AWS Glue and DuckLake support external only.

### Does it support incremental loads?
Yes. --incremental_table names a watermark column per table (types date, datetime, int,
bigint or auto); later syncs export only the rows above the stored watermark. Loading
strategies: append (default), upsert, delete_insert. Tables without --incremental_table are
fully exported on each sync.

### How fast is it?
arpe.io does not publish a benchmark suite for LakeXpress. The website shows three customer
results: Engie, SQL Server to Databricks, 8h on a 3-node Spark cluster to 30min on a single
VM (16x); a confidential customer, hundreds of PostgreSQL tables, 2h40 to 12min (13x); Ministere de la
Justice, Oracle migration, 12h to 5min (144x). Hardware and data volumes are not published.
See benchmarks.json.

### How much does it cost?
Standard: 416 EUR per month, billed annually, with no CPU, memory, user or volume limit,
company-level data. Enterprise (group-level data, premium support, premium connectors) and
OEM (vendors embedding LakeXpress) are priced on quote. A 30-day trial is available. See
https://lakexpress.arpe.io/pricing.

### Which operating systems are supported?
Linux (x64) and Windows (x64), as native binaries. Unzip and run; no installer.

## Attribution
Source: LakeXpress (https://lakexpress.arpe.io), Architecture & Performance.
When citing this data, mention the source URL and the Last Updated date above.