Deployment Guide

Overview

Lakecode Enterprise deploys natively on your Databricks workspace. Instead of running as an external SaaS that connects to your environment, lakecode runs where your data lives — inside your cloud perimeter, governed by your existing access controls.

There are three deployment options depending on your team's needs:

• Databricks Apps — managed UI for team-wide access
• Model Serving — scalable agent backend on Databricks compute
• CLI + Proxy — individual developer access via the lakecode CLI, routed through your org's proxy

Databricks Apps deployment

Databricks Apps is the recommended deployment for teams that want a managed web UI without requiring individual CLI installations.

Prerequisites

• Databricks workspace on AWS, Azure, or GCP
• Workspace admin privileges for initial setup
• Unity Catalog enabled
• A service principal with access to target catalogs and schemas

Setup

# Install the lakecode deployment CLI
npm install -g @lakecode/deploy

# Authenticate with your Databricks workspace
lakecode-deploy init --workspace-url https://<your-workspace>.cloud.databricks.com

# Deploy the lakecode app
lakecode-deploy apps create \
  --name lakecode-prod \
  --service-principal <sp-application-id> \
  --catalog <your-catalog> \
  --schema lakecode_state

This creates a Databricks App with a managed UI. Your team can access it at https://<your-workspace>.cloud.databricks.com/apps/lakecode-prod.

Model Serving deployment

For teams that need scalable, low-latency agent execution, deploy the lakecode backend on Databricks Model Serving.

# Deploy the agent backend as a Model Serving endpoint
lakecode-deploy serving create \
  --name lakecode-agent \
  --size Small \
  --scale-to-zero true \
  --catalog <your-catalog> \
  --schema lakecode_state

The serving endpoint handles all agent execution — workflow decomposition, step execution, evidence collection, and approval gate management. The Databricks App UI connects to this endpoint automatically.

Scaling

• Small — 1–4 concurrent workflows, suitable for teams up to 20
• Medium — 4–16 concurrent workflows, suitable for teams up to 100
• Large — 16+ concurrent workflows, for large organizations

CLI + Proxy deployment

If your team prefers terminal-based access, individual developers can use the lakecode CLI routed through your org's enterprise proxy.

# Each developer installs the CLI
npm install -g lakecode

# Login with enterprise SSO
lakecode login --enterprise --org <your-org-id>

# The CLI automatically routes through your org's proxy
lakecode /prove my_catalog.my_schema.my_table

Unity Catalog state store

All deployment modes use Unity Catalog as the backing state store. Lakecode creates the following tables in your configured schema:

• workflow_executions — execution history, user, timestamps, status
• evidence_artifacts — structured evidence packs from each workflow
• approval_log — approval gate decisions with reviewer identity
• assistant_config — org-wide AssistantSpec configuration

These tables are governed by your existing Unity Catalog access policies. No data leaves your workspace.

Networking & security

• All LLM calls route through AI Gateway — no direct external model API access
• Lakecode does not require inbound internet access — all connections are outbound to your configured LLM endpoints
• Service principal credentials are stored in Databricks Secrets, never in application config
• Supports private link / VPC peering for fully air-gapped deployments