Drupal TMGMT AI Translation: Which Module to Choose
Drupal has more than one way to put an LLM behind your translations, and the module names are easy to confuse. The short version: use AI Translate for one-off translations on a single node, use AI TMGMT when you need bulk jobs with a review queue, skip tmgmt_openai because it is obsolete, and look at a commercial connector or a hosted service when you do not want to run the AI stack yourself.
This guide compares those routes, walks through the AI TMGMT setup, lists the pitfalls to check before a big batch, and shows how to estimate cost. It is written for Drupal 10 and 11 site builders who already know Content Translation and the Language module.
Disclosure: we build LATW Cloud, one of the options covered below. Module details come from the drupal.org project pages as read on 16 November 2026; we describe them from documentation and have not tested every module hands-on.
Key Takeaways
- AI Translate works inside the node’s Translate tab. AI TMGMT adds jobs, bulk selection and TMGMT’s review step.
- AI TMGMT had no stable release as of November 2026 (1.0.0-beta6), so test it on staging first.
- tmgmt_openai is marked obsolete on drupal.org. Its page points to AI TMGMT as the replacement.
- Estimate words and cost before you run a whole-site job. Per-token prices and prompt length decide the bill.
- A hosted option such as LATW Cloud trades control of the AI stack for less setup.
Which Drupal AI translation route fits which job?
Pick by the job, not by the module name. The table summarizes what each option is for, based on the drupal.org documentation.
| Route | Best for | Review step | Stability (Nov 2026) |
|---|---|---|---|
| AI Translate | One node at a time, from the Translate tab | You edit and save the translation yourself | 1.4.4 stable |
| AI TMGMT | Bulk jobs across many nodes, using your own AI provider | TMGMT job review and accept | 1.0.0-beta6, no stable release |
| tmgmt_openai | Nothing new: marked Unsupported and Obsolete | TMGMT jobs | Replaced by AI TMGMT |
| Commercial connectors (Lilt, Smartling, Lara) | Teams already paying for that vendor | Vendor workflow plus TMGMT | Lilt 1.7, Smartling 9.28, Lara 2.0.0-beta7 |
| Hosted service (TMGMT LATW Cloud) | Background jobs without running the AI stack | Orders placed in the vendor panel | See below |
TMGMT itself (8.x-1.19, released 27 August 2026) is the job framework underneath. It ships with XLIFF and HTML export and import, a local translator, and machine-translation connectors for DeepL, Google Translate and Microsoft Translator (TMGMT project page). The AI options plug into it as additional translators.
AI Translate: the inline option
AI Translate “adds one-click, AI-powered translations to Drupal’s content translation system” by taking over the Translate tab (AI Translate project page). It needs core Content Translation plus the AI module with a provider configured for the translation operation.
Per its documentation it offers field-aware extraction, referenced content up to a set depth, a model and prompt per language, editable prompts stored as configuration, and Drush support. Setup is three steps:
composer require drupal/ai_translateanddrush en ai_translate.- Grant the “Create AI translation” permission to the roles that should use it.
- Set prompts, models and reference depth under Administration > Configuration > AI > AI Translate Settings.
AI Translate used to live inside the AI module. The project page states it was removed from there in AI 1.4.0, so older tutorials that tell you to enable it from the AI module are out of date. Current versions listed are ai_translate 1.4.4 (stable, Drupal ^10.4 || ^11) and AI 1.5.2 (stable, ^10.5 || ^11.2, released 28 October 2026).
Use it when editors translate a handful of pages and want to see the result immediately. For a hundred nodes, clicking through tabs stops being practical.
AI TMGMT: bulk jobs with a review queue
AI TMGMT is “an AI translator plugin for the Translation Management Tools (TMGMT) project” built on the AI module (AI TMGMT project page). It lets you select one or many nodes, creates TMGMT jobs, and keeps TMGMT’s review process. It works with Drupal ^10 || ^11.
One status note matters for planning: the page says “There are currently no supported stable releases.” The latest release at the time of writing is 1.0.0-beta6 from 11 June 2026. That is workable for a staging run, but check the release notes before you rely on it for client production work.
Setup steps
These follow the project page.
- Install the module and its dependencies.
composer require 'drupal/ai_tmgmt:^1.0@beta'. The AI module needs the Key module, which is its only listed requirement (AI project page). Note the AI module 1.5.x line requires Drupal 10.5 or later. Install TMGMT and enable at least one source, such as Content Entity Source. - Configure an AI provider at
/admin/config/ai/settings. The AI project page says it supports 95+ providers, and the AI TMGMT page names OpenAI and Ollama. Store the API key through the Key module, not in code. - Enable the TMGMT AI Provider at
/admin/tmgmt/translators. - Run a test job. Go to
/admin/tmgmt/sources, select a node, choose the target language and request a translation. - Review and accept the job in TMGMT. Nothing replaces the target-language node until you accept it, which is the main reason to choose TMGMT over the inline tab for client work.
If you are leaving tmgmt_openai, the project page’s migration path is: install ai_tmgmt, check the new provider, finish any open tmgmt_openai jobs, disable the old provider, then uninstall tmgmt_openai (tmgmt_openai page).
Connecting AI TMGMT to AI Translate
AI Translate documents a “Framework mode” for use as the engine behind other tools such as AI TMGMT. AI TMGMT’s own description mentions per-language model and prompt via the AI Translate submodule. The two pages disagree on one detail: the AI TMGMT page still says the integration uses the submodule of the AI module, while AI Translate is now a separate project. Install ai_translate explicitly if you want the per-language prompts.
Pitfalls to check before a bulk run
The first two come from the documentation and the issue queue; the rest are general checks. Test each on a staging copy with five or ten nodes.
- Jobs processed twice. The AI TMGMT page recommends a patch for TMGMT issue #3198609, “Same Job processed/checkout twice by the JobCheckoutManager”. Without it, the same job can be processed twice, which would mean paying your provider twice for the same text.
- Prompt mismatch. Issue 3533689 reports that the “Use the configuration from the AI Translate module” setting is not respected: AI TMGMT kept its simple prompt (“Translate from German into English language”) instead of the detailed AI Translate prompt, visible in the AI logs. We could not read the issue’s current status, so check it before you assume your per-language prompts apply. The AI logs show the prompt actually sent.
- HTML and field types. Check how rich-text fields, paragraphs, link fields and metatags come back. Translate a node that uses every field type on your site and compare the result field by field.
- Large batches and timeouts. Run jobs through queues or Drush rather than one long browser request, and start with a small batch to see how your provider’s rate limits behave.
- Glossary and brand tone. A plain provider call has no built-in glossary. Put protected terms and tone instructions in the prompt, then spot-check that product names survive.
- API key cost control. Set a spending limit at the provider and read the AI logs after the first job.
How to estimate cost before translating a whole site
Cost is words times target languages times price per token, plus prompt overhead. A rough method:
- Export or count the words of the content you will translate. Count each target language separately: 1,000 words into three languages is 3,000 translated words.
- Run a five-node pilot and read the token usage in the AI logs. Divide by the words you sent to get tokens per word, including your prompt.
- Multiply by your total words and your provider’s current per-token price.
Model choice moves the result more than anything else. Our benchmark of six models against DeepL and Google Translate shows the spread: see what AI translation actually costs. If your provider offers a batch API, Batch API cuts whole-site cost explains when the discount is worth the wait.
Commercial TMGMT connectors
If your team or client already pays for a translation platform, a connector keeps everything in one workflow. As listed on drupal.org on 16 November 2026:
- Lilt:
tmgmt_lilt1.7 (May 2025) needs TMGMT and a Lilt API key (project page). - Smartling:
tmgmt_smartling9.28 (October 2026) needs TMGMT, the TMGMT Extension Suite and an active Smartling subscription (project page). - Lara Translate:
tmgmt_laratranslate2.0.0-beta7 (November 2026), labelled “Minimally maintained” on its page (project page). - Transifex: we did not find a TMGMT connector for it on the TMGMT provider list or ecosystem pages. We did not run a full drupal.org search, so one may exist.
We have not tested these connectors. Check each vendor’s current pricing and plan limits directly; they change and we do not quote them here.
When you do not want to run the AI stack yourself
Running AI TMGMT means you maintain the AI module, a provider account, keys, prompts and job retries. That is fine for teams that want control. If you would rather hand that part off, a hosted service is the alternative.
LATW Cloud is one such service, and it has a Drupal module, TMGMT LATW Cloud (AI Translation). Facts from the module page on latw.ai:
- Works with Drupal 10.6 and 11, PHP 8.1 or later, TMGMT 1.18 or later with tmgmt_content, and core Content Translation.
- Install with
composer require drupal/tmgmt_latw && drush en tmgmt_latw, then go to/admin/tmgmt/latwand click “Connect with LATW”. - Orders are placed in the LATW panel, not in the TMGMT job form.
- Your site must be reachable from the internet, with
POST /latw/v1/cloud/*allowed through the firewall. - It is AI-only; there is no human translator service. Uninstalling removes the connection, and translations already saved stay on your site.
Pricing, as of November 2026 on latw.ai/cloud: the Free plan has no monthly fee, and your first site gets 10,000 words on the Economy quality class. After that you pay per word through Credits: Economy is 50 Credits per 1,000 words, Standard 1,000 and Premium 5,000, per target language. On the Free plan a pack of 10,000 Credits costs $12. By that arithmetic a 50,000-word site into one language on Economy is 2,500 Credits, or about $3.00 at the pack price. You see an estimate before each job and a job never costs more than that estimate. Paid plans add things such as batch mode at half the Credits and bring-your-own-key through OpenRouter or OpenAI.
LATW Cloud also supports glossaries for terms that must stay untranslated. For agencies running several client sites, managing client sites from one account is covered separately. You can try it on the free allowance first.
Trade-offs to weigh: you hand your content to a third party (the plan page states text is deleted 7 days after a job finishes and hosting is in the EU), and you place orders in LATW’s panel instead of the Drupal job form. If review inside TMGMT is a hard requirement, AI TMGMT fits better.
Frequently asked questions
Is AI TMGMT production ready?
It has no stable release as of November 2026; the latest is 1.0.0-beta6. Whether a beta is acceptable is your call per project; test on staging and apply the recommended patch.
Can I keep using tmgmt_openai?
Its drupal.org page marks it Unsupported and Obsolete and says development moved to AI TMGMT. Plan the migration steps above.
Do I need AI Translate if I use AI TMGMT?
Not to run basic jobs. You need it if you want per-language models and prompts, and you should check issue 3533689 for how the prompt setting behaves.
Which route is cheapest?
The modules themselves are free to install, except that commercial connectors need a paid vendor account. For AI TMGMT the cost is your provider’s AI usage, so the model you pick and your prompt length matter most. Run a pilot, read the logs and extrapolate.
Where can I read about other platforms?
Related guides: the same cost logic on PrestaShop or Webflow localization alternatives, and AI translation services compared.
What to do next
Pick the route by job size. Single pages: AI Translate. Bulk jobs with review and your own provider: AI TMGMT on staging, with the double-processing patch and a pilot batch. Hosted background jobs: try LATW Cloud on the free allowance. Whichever you choose, run five nodes first and read the output field by field before you queue the whole site.