Dextalo docs

# Clone repository
git clone https://github.com/eqho-dev/sls-cli.git
cd sls-cli

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install CLI dependencies
pip install -r cli/requirements.txt

# Verify
python -m cli.dxt_cli --help

# If your installation includes a template
cp .env.example .env

# Otherwise create .env manually, then edit it with your credentials
nano .env

# Optional Google Cloud settings for dextalo predict
GOOGLE_APPLICATION_CREDENTIALS=/path/to/your/service-account-key.json
PROJECT_ID=your-google-cloud-project-id
LOCATION=us-central1
MODEL_ID=general/nmt

# OpenRouter / AI workflow
OPENROUTER_API_KEY=your_openrouter_api_key
AI_MODEL=google/gemini-3-flash-preview
AI_STYLE_GUIDE_MODEL=openai/gpt-5-mini
AI_TRANSLATE_MODEL=google/gemma-3-27b-it
AI_EDITING_MODEL=openai/gpt-5-mini
AI_QA_MODEL=openai/gpt-5-mini
AI_REPAIR_MODEL=google/gemini-3.1-flash-lite-preview

# Database (required for TM features)
TM_DB_PROVIDER=local
TM_DB_HOST=localhost
TM_DB_PORT=5432
TM_DB_NAME=predict
TM_DB_USER=postgres
TM_DB_PASSWORD=your_password

GOOGLE_APPLICATION_CREDENTIALS=/path/to/your/service-account-key.json

# macOS / Linux
export GOOGLE_APPLICATION_CREDENTIALS="/path/to/key.json"

# Windows PowerShell
$env:GOOGLE_APPLICATION_CREDENTIALS="C:\path\to\key.json"

# Create a working directory
mkdir -p ~/my-translation-project
cd ~/my-translation-project

# Initialize the workspace
dextalo init

# Put source XLIFF in source/
# Put TMX in memories/
# Put glossary files in glossaries/
# Put reviewed QA workbooks in qa_workbooks/ when needed

# Launch the interactive app
dextalo

# Then choose:
#   Languages           set the source/target pair once
#   Update TM           optional, if you have TMX files
#   Pipeline            analyze-content → translate → edit → qa
#   Apply QA            optional targeted fixes from QA findings
#   Grade               optional evaluation step
#   Restore             optional document export
#
# Predict remains available as a separate standalone MT step.

my-translation-project/
├── source/            # Input XLIFF files
├── target/            # Translated output files
├── temp/              # Processing intermediates
├── grading/           # Reference files for quality evaluation
├── glossaries/        # Terminology CSV files
├── memories/          # Translation memory files (TMX)
├── linguist_profiles/ # Personalization samples and learned QA guides
├── qa_workbooks/      # Reviewed QA Excel workbooks for interactive learning
├── convert/           # Documents to convert to XLIFF
├── deliverable/       # Restored documents (DOCX/PPTX/TXT)
├── logs/              # Log files
└── .dxt               # Workspace marker file

dextalo

─── Translation Workflow ───
  AI Translate          AI translation with TM and glossary context (RAG)
  Edit                  AI post-editing pass
  QA                    Run structured localization QA validation
  Apply QA              Revise target files using QA findings
  Grade                 Evaluate translation quality
  Pipeline              Run complete workflow (analyze-content → translate → edit → qa)
  Repair                Detect segment-shift corruption in edited XLIFF files
  Apply Repair          Apply fixes recorded in the repair report

─── Settings ───
  Change Source Folder  Set source file directory
  Change Target Folder  Set target file directory
  Workspace Prefix      Toggle workspace/ folder prefix
  Translation Memories  Select active TM from database
  Glossaries            Select active glossary CSV
  Style Guide           Select active style guide TOML
  Domains               Select active domain guidance
  Personalization       Select or generate linguist personalization guide
  Languages             Set source and target language

─── Analysis & TM ───
  Analyze File          Analyze XLIFF files for metrics
  Generate Style Guide  Generate AI-powered style guide
  Update TM             Import TMX files to database
  Embed TM              Generate vector embeddings for TM segments

─── Tools ───
  Convert               Convert file to XLIFF
  Restore               Restore XLIFF to XLSX, DOCX, or other format

─── System ───
  Version               Show version information
  Config                View configuration
  Check Update          Check for available updates

# 1. Initialize a workspace
dextalo init

# 2. Add files to the workspace folders
#    source/             XLIFF files to translate
#    memories/           TMX files
#    glossaries/         glossary files
#    linguist_profiles/  personalization samples if you want voice guides
#    qa_workbooks/       reviewed QA Excel workbooks
#    grading/            reference files for scoring

# 3. Launch the interactive app
dextalo

# 4. In the menu, choose:
#    Languages         set the source and target pair once
#    Domains           optional domain prompt layer
#    Personalization   optional linguist style guide
#    Workspace Prefix  optional, if folders should live under workspace/
#    Update TM         optional
#    Pipeline          analyze-content → translate → edit → qa
#    Apply QA          optional targeted fixes
#    Repair            optional corruption detection after editing
#    Grade             optional evaluation
#    Restore           optional document export

# 1. Initialize workspace
dextalo init

# 2. Import TM
dextalo update-tm

# 3. Run the current audited pipeline
dextalo pipeline

# 4. Apply QA-driven revisions if needed
dextalo apply-qa

# 5. Detect and fix segment-shift problems if needed
dextalo repair
dextalo apply-repair

# 6. Quality grading
dextalo grade reference.xlf candidate.xlf

# 1. Run the normal translation flow
dextalo pipeline

# 2. Restore to Excel if the reviewer wants a workbook
dextalo restore

# 3. Human QA edits the restored workbook
#    Active sheet must contain: Source | Target | Edit
#    Edit = human-corrected target text

# 4. Copy the reviewed workbook into qa_workbooks/

# 5. Open the interactive app
dextalo

# 6. Choose QA
#    First picker: select XLIFF files in target/
#    Second picker: optionally select one workbook in qa_workbooks/

# Direct CLI alternative
dextalo qa --reviewed-workbook "deliverable/bkk air batch 26 new_restored.xlsx"

cd ~/my-translation-project
dextalo init

TM_DB_PROVIDER=local
TM_DB_HOST=localhost
TM_DB_PORT=5432
TM_DB_NAME=predict
TM_DB_USER=postgres
TM_DB_PASSWORD=your_password

# Create database
createdb -U postgres predict

# Apply TM schema
psql -U postgres -d predict < cli/database/create_translation_memory_schema.sql

TM_DB_PROVIDER=supabase

# Full connection URL (recommended)
SUPABASE_DB_URL=postgresql://postgres.[REF]:[PASS]@aws-0-[REGION].pooler.supabase.com:5432/postgres

# Or individual settings
SUPABASE_HOST=aws-0-[REGION].pooler.supabase.com
SUPABASE_PORT=5432
SUPABASE_USER=postgres.[REF]
SUPABASE_PASSWORD=your_password
SUPABASE_DATABASE=postgres

# macOS
brew install pgvector

# Ubuntu / Debian
sudo apt install postgresql-16-pgvector

psql -U postgres -d predict < cli/database/migrate_add_vector_search.sql

# Dry run first
python cli/database/backfill_embeddings.py --dry-run

# Generate for all TM sets
python cli/database/backfill_embeddings.py

# Specific TM set
python cli/database/backfill_embeddings.py --tm-set "My_TM_Set"

# Use local model
python cli/database/backfill_embeddings.py --provider local

SELECT COUNT(*) AS total,
       COUNT(embedding_openai) AS with_embeddings
FROM translation_memory.tm_segments;

TM_EMBEDDING_PROVIDER=openai
OPENAI_API_KEY=sk-your-key-here

TM_EMBEDDING_PROVIDER=local

psql -U postgres -d predict < cli/database/optimize_tm_performance.sql

AI_EDITOR_TEMPERATURE=0.1  # Recommended for translations

AI_EDITOR_INCLUDE_SOURCE=true

# Case-insensitive matching: "ON" matches "On", "on"
GLOSSARY_IGNORE_CASE=true