Skip to content

Configuration (leai.yml)

The leai.yml configuration file controls all aspects of extraction, filtering, storage paths, GitOps version control, and AI integrations for LEAI.

📄 Complete Annotated Example

# ==============================================================================
# LEAI Configuration
# ==============================================================================

# 1. Oracle Connection (DSN)
# Supports environment variable interpolation via ${VARIABLE}
dsn: "oracle://${DB_USER}:${DB_PASS}@${DB_HOST}:1521/${DB_SERVICE}"

# 2. Target Schemas
# Can be a list of schema names or "ALL" (requires DBA / SELECT ANY DICTIONARY)
schemas:
  - HR
  - SALES

# 3. Pipeline Paths & Logs
rawPath: "./raw"                  # Technical snapshots in JSON
annotationsPath: "./annotations"  # Business annotations in YAML
docPath: "./docs"                  # Final Markdown documents
updates_log_path: "./logs/updates" # Update audit logs and manifests (latest.json, latest.md)
generate_update_log: true         # Whether to generate update audit logs in leai update

# 4. Inclusion & Exclusion Filters (SQL LIKE Wildcards)
include:
  - EMPLOYEES
  - ORDERS_%
exclude:
  - BIN$%                         # Oracle Recycle Bin tables
  - SYS_%

# 5. Object Categories
object_types:
  - tables
  - views
  - mviews
  - procedures
  - functions
  - packages
  - triggers
  - synonyms

# 6. AI Settings (LLMs for Auto-Enrichment, Chat, and Subagents)
ai:
  default_provider: "ollama"      # ollama | local | openai | gemini | anthropic | deepseek | qwen | kimi | grok
  temperature: 0.2                # Global default temperature (0.0 to 1.0)
  timeout: 300.0                  # Global default request timeout in seconds
  max_history_turns: 15           # History context memory window in chat turns
  max_agent_iterations: 10        # Maximum reasoning tool iterations per turn
  max_subagent_iterations: 5      # Maximum iterations for specialized subagents
  providers:
    ollama:
      base_url: "http://localhost:11434/v1"
      model: "qwen2.5-coder:latest"
      temperature: 0.1
      timeout: 300.0
      num_ctx: 32768              # Context window size in tokens to prevent truncation on large DDLs
      keep_alive: "1h"            # Keep model loaded in memory on Ollama server
    local:
      base_url: "http://localhost:1234/v1" # e.g. LM Studio, vLLM, LocalAI
      model: "qwen2.5"
      temperature: 0.1
      num_ctx: 32768
      max_tokens: 4096
      top_p: 0.95
    openai:
      api_key: "${OPENAI_API_KEY}"
      model: "gpt-4o-mini"
      temperature: 0.2
      timeout: 120.0
      max_tokens: 4096
    gemini:
      api_key: "${GEMINI_API_KEY}"
      model: "gemini-2.5-flash"
    anthropic:
      api_key: "${ANTHROPIC_API_KEY}"
      model: "claude-3-5-sonnet-20241022"

# 7. Git / GitLab / GitHub Synchronization (GitOps)
git:
  enabled: false                                 # Enables leai git commands and /git
  remote_url: "${GIT_REMOTE_URL}"                # Remote repository URL
  branch: "main"                                 # Tracking branch
  author_name: "LEAI Bot"                        # Commit author name
  author_email: "leai@company.com"               # Commit author email
  auto_sync: false                               # Automatic push after extract/compile
  tracked_paths:
    - "annotations"
    - "docs"
    - "raw"
    - "leai.yml"

# 8. Distributed Storage / Object Storage (SeaweedFS / S3)
storage:
  seaweedfs:
    enabled: false                                 # Automatically routes operations to S3
    endpoint_url: "http://localhost:8333"          # SeaweedFS or MinIO S3 gateway
    bucket: "leai"                                 # S3 bucket name
    access_key: "${SEAWEEDFS_ACCESS_KEY}"
    secret_key: "${SEAWEEDFS_SECRET_KEY}"
    region_name: "us-east-1"
    raw_prefix: "raw"                              # Folder prefix for JSON snapshots
    annotations_prefix: "annotations"              # Folder prefix for YAML annotations
    auto_create_bucket: true                       # Creates bucket if missing
    no_cache: false                                # Local disk cache or pure remote
    incremental: true                              # SHA-256 hash deduplication

# 9. Interface Language & Localization
language: "en-US"                                  # "en-US" (canonical default) or "pt-BR"

# 10. Automatic PyPI Update Check
update_check: true                                 # true (default) or false

🔑 Supported DSN Formats

LEAI supports various ways to declare connection strings:

URL Syntax (Standard)

dsn: "oracle://user:password@host:1521/service_name"

EZCONNECT Syntax (Native Oracle)

dsn: "user/password@host:1521/service_name"

Full TNS Descriptor (for TCPS, Wallets, or Oracle Cloud / Autonomous DB)

dsn: "user/password@(DESCRIPTION=(ADDRESS=(PROTOCOL=TCPS)(HOST=db.example.com)(PORT=1522))(CONNECT_DATA=(SERVICE_NAME=my_service)))"

🎯 Object Filtering

You can use the include and exclude filters to selectively target database entities:

  • %: Matches zero or more characters (e.g. TB_% matches any object starting with TB_).
  • _: Matches exactly one character.

[!TIP] If include is left empty, LEAI will extract all objects matching object_types that are not explicitly matched by exclude.

🌐 Internationalization & Language (language)

LEAI provides a built-in native internationalization engine (leai/i18n) with English as canonical default and parity bilingual catalogs (en-US and pt-BR):

language: "en-US" # or "pt-BR" (default: "en-US")

What the Language Setting Affects:

  1. Terminal & CLI: Formatted status panels, tabular outputs, progress spinners, and pipeline execution summaries.
  2. Interactive TUI Session: Command responses (/git status, /help, /rule, /copy, /model).
  3. AI Prompts & Responses: When configured to pt-BR, the system prompt instructs the LLM to enrich empty column descriptions and business rules in Brazilian Portuguese (while preserving technical SQL identifiers). In en-US, generation is performed strictly in English.
  4. Compiled Markdown Documentation: Documents generated under docs/ use localized headers (## Overview, ## Columns, ## Business Rules vs. ## Visão Geral, ## Colunas, ## Regras de Negócio).
  5. Web Studio: Web user interface labels, modals, loading spinners, and real-time language switching via REST API (/api/config).

Resolution Precedence

LEAI determines the active locale according to the following strict priority: 1. CLI Flag: --lang <locale> or -L <locale> (e.g. leai --lang en-US ask "what are the main sales tables?") 2. Environment Variable: LEAI_LANG or LEAI_LANGUAGE (e.g. export LEAI_LANG=en-US) 3. Configuration File: language: key in leai.yml 4. Canonical Fallback: en-US

Quick Initialization with Language

To generate a leai.yml pre-configured and documented in Portuguese: 

leai init --lang pt-BR
Or in English (default): 
leai init --lang en-US

🔄 Automatic Update Check (update_check)

By default, LEAI performs a lightweight non-blocking query to PyPI on startup to verify whether a newer release is available.

update_check: true # true (default) or false

How to Disable Update Checking:

  1. In leai.yml: Set update_check: false.
  2. Via Environment Variable: Export LEAI_NO_UPDATE_CHECK=1 or LEAI_NO_UPDATE_CHECK=true.
  3. Via Command Line (Global Flag): Pass --no-update-check to any command: 
    leai --no-update-check extract
leai --no-update-check chat

📋 Update Audit Logs (updates_log_path & generate_update_log)

When running leai update, LEAI can automatically record a manifest and report of all objects extracted and modified in Oracle:

updates_log_path: "./logs/updates" # Output directory for log files (Default: ./logs/updates)
generate_update_log: true         # Generates JSON and Markdown audit logs (Default: true)

Generated Files per Run:

  • update_YYYYMMDD_HHMMSS.json & latest.json: Structured manifest with UTC timestamp, search window, modifier (last_modified_by), Oracle timestamp (last_ddl_time), and sync counts.
  • update_YYYYMMDD_HHMMSS.md & latest.md: Readable report with formatted tables for VS Code or GitHub.
  • SeaweedFS / S3 Synchronization: If --seaweed or S3 storage is enabled, files are automatically synced to the bucket under logs/updates/.