Skillget

localization-os

mcp
Guvenlik Denetimi
Uyari
Health Uyari
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 8 GitHub stars
Code Gecti
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

Localization OS: open-source, self-hosted, MT-first translation management system

README.md

Localization OS

Own your localization stack. Localization OS is an open source, self-hosted, MT-first translation management
system (TMS). Upload a file or connect a CMS, let machine translation do the first pass with your translation memory,
glossary and style guide, check quality automatically, post-edit only what needs a person, and get the file back in
its original format.

It runs on your own machine or server, on cloud AI engines or fully local models, and nothing leaves your network
unless you tell it to.

Quick start (about 5 minutes)

You need Python 3.11 or newer (CI and the container use 3.13).

git clone https://github.com/patois-labs/localization-os.git
cd localization-os
python3 scripts/bootstrap.py     # macOS / Linux
python scripts/bootstrap.py      # Windows (or: py scripts\bootstrap.py)

bootstrap.py creates a .venv, installs the pinned runtime dependencies AND the developer tools (test runner, browser automation, audit tools), so the .venv is about 600 MB, and copies .env.example to .env. Then start it:

# macOS / Linux
.venv/bin/python run.py
# Windows
.venv\Scripts\python.exe run.py

Open http://127.0.0.1:8000 (set PORT and HOST in .env to change it) and sign in with admin / admin (you are asked to choose a new password). The default
engine is the built-in mock engine, so everything works offline and free with obviously fake translations. Add a
key for Claude, Gemini or OpenAI in .env, or point it at a local model server (LM Studio, Ollama, vLLM), and
translations become real with no code changes.

Prefer containers? A local trial runs with docker compose up on SQLite. Before an image serves anyone else, run
the image scan gate in docs/DEPLOY.md, which also covers Postgres, TLS, Podman and the first-admin
setup wizard.

What it does

Translation

  • Configurable workflows: translate, quality check, fix and re-check loops, human review steps, pause and resume.
  • Describe a workflow in plain language and it drafts the steps for you to review.
  • Translation memory (TMX, exact and in-context matching, locks, a word leverage report), glossaries with do-not-translate and required terms, style guides,
    content types and context packages, all sent to the model and verified on the way back.
  • Automated quality checks: MQM-style scoring, tags, numbers, terms, placeholders, refusals.
  • A post-editing editor for the segments that need a person.
  • Engines behind one switch: Claude, Gemini, OpenAI, DeepL, Google, Microsoft, Amazon, local self-hosted models and any
    OpenAI-compatible endpoint, with failover and a spend cap.

Content in and out

  • More than 80 file formats (Office, PDF, HTML, XML/DITA, Markdown, InDesign, subtitles, software strings, XLIFF and
    CAT packages, images) returned in the same format and layout. An untouched file comes back byte for byte.
  • Three project types: documents, string keys (with a delta API for CI), and projects connected to a source system.
  • 28 connectors, including Airtable, Jira, Strapi, Contentful, HubSpot, WordPress, Shopify, Zendesk, Confluence,
    GitHub, GitLab, SharePoint and Google Docs. Translations go back as each system's own language version; the source
    is never overwritten.
  • External vendors: send work to a human vendor and receive a signed return.

Security and administration

  • Role-based access, project membership, SSO (OIDC, SAML, LDAP / Active Directory), SCIM provisioning, MFA.
  • A tamper-evident audit chain, an LLM Ledger of every model call, GDPR export and erasure, PII redaction, encryption
    at rest.
  • Reports, a dashboard, a reviewer worklist, comments and queries, data export, retention rules.

Built for agents

  • Every feature is an API first. A full REST API, a headless CLI with more than 480 commands, and an MCP server (72 task-shaped tools) so AI
    agents such as Claude can run projects, workflows, translation memory searches and connector syncs.

Paid add-ons and hosting

The core in this repository is complete and free. Patois Labs also offers paid add-ons for specialist needs (for
example an agent-based advanced workflow tier, an evaluations lab, semantic translation memory and enterprise
infrastructure adapters), plus managed hosting, private AI on your hardware or ours, support, custom integrations
and procurement paperwork. What stays free forever is written down in STEWARDSHIP.md. Details:
https://patoislabs.com.

Documentation

Contributing

Contributions are welcome. Read CONTRIBUTING.md (setup, tests and the DCO sign-off), the
code of conduct, and AGENTS.md if you work with an AI coding assistant. Report
security issues privately as described in SECURITY.md, never in a public issue.

License and trademarks

Copyright 2026 Patois Labs, LLC. Licensed under the Apache License, Version 2.0; see LICENSE and
NOTICE. Third-party components keep their own licenses, listed in
THIRD_PARTY_NOTICES.md. The Apache license does not grant rights to the Patois Labs names
and logos; see TRADEMARKS.md.

Yorumlar (0)

Sonuc bulunamadi