Custom.MT

What is Custom.MT?

Custom.MT is an enterprise AI translation and localization orchestration platform developed by CMT Language Technology s.r.o.; it serves localization teams, language service providers, and companies that require multi-language content processing.

It does not offer just a single translation model; instead, it allows users to combine machine translation, generative AI, terminology, prompts, quality checks, and automatic editing steps in the console, and then integrate this workflow into existing translation systems.

Main functions

• Centralized management of multiple models: It allows for the unified configuration of over 30 different neural machine translation and large language models; it also enables integration with proprietary, custom, or open-source models, thereby reducing the need to switch between various supplier platforms.
• Translation workflow template: Users can combine translation, terminology management, pre-editing, post-editing, quality checking, and rollback functions based on language pairs, content areas, and project requirements, and reuse them across different projects.
• AI Terminology: It is possible to include corporate terminology in the template, so that the model’s outputs are more in line with product names, industry-specific terms, and existing linguistic standards; the scope of these terms depends, however, on the specific engine and workflow configuration.
• Prompt Studio: The Advanced and higher plan versions allow users to create custom prompts in order to control the tone, style, and translation instructions; these same rules can be applied to connected CAT or TMS tools.
• Best Engine: After uploading manually translated samples, multiple engines can be used to process the same content; BLEU, COMET, WER, edit distance, and sample excerpts can be compared to help select the most suitable model.
• Automatic front-and-back editing: The workflow allows for the cleaning of the input text before translation, or it can use generative AI to revise the output of machine translation; multi-stage processing increases the consumption of characters or tokens.
• Quality estimation and QA: Premium packages provide quality estimates and quality inspection data, which are useful for determining the priority of manual reviews; however, automated scores cannot replace the scrutiny of critical content by professional translators.
• Tag and short code handling: It is possible to retain, remove, or protect tags, placeholders, and short codes in software strings, thereby reducing the risk of format disruption after translation.
• Usage and team management: It is possible to view the amount of processing carried out by date and by model; Compliance Edition also provides team dashboards and an enterprise translation portal.

Input, processing, and output

The process from trial use to production use

1. Register a new account to access a 7-day free trial; first determine the language pair, content type, monthly character limit, and data sensitivity level.
2. Add managed engines to the console, or connect to keys from your own providers such as DeepL and OpenAI; do not place production keys in clients that are publicly accessible.
3. Create a workflow template by selecting the primary engine, backup engine, terms, prompts, tag rules, as well as the nodes for pre-editing, post-editing, and quality checking.
4. The output is tested using real but anonymized small-scale samples, and its performance is assessed both with the Best Engine metric and through manual review.
5. Generate a console API key, or install the appropriate connector in tools such as Trados, memoQ, Smartling, etc. and select a template.
6. After going live, the monitoring console processes characters, supplier-specific characters or tokens, as well as the quality score and the amount of manual modifications, thereby adjusting the templates on a continuous basis depending on the project.

Prices and billing structure

Custom.MT employs a two-tier pricing system: first, a subscription fee for the console is paid, and then a fee is charged for the actual translation engine usage. The public prices are listed in euros and do not include value-added tax.

The cost for translation is charged separately.

• Traditional machine translation typically charges per character, while generative AI usually calculates the cost based on the input and output tokens separately.
• Visible text, tags, markers, spaces, and the content of requests can all be included in the character count; empty segments may also be charged as if they consisted of at least one character each.
• Translation into multiple target languages is measured separately; translating the same text into three languages results in roughly three separate processing tasks.
• Prompt words, terms, and multi-stage editing before and after increase the amount of data that needs to be processed by the model, and the actual cost of complex workflows can be higher than the number of characters in the original text.
• When the capacity of the console is exceeded, the service is not interrupted immediately; instead, charging continues at the overage rate specified for that plan.

Cancellation and refund

New users have a 7-day trial period; after that, their accounts are suspended until an upgrade is made. No refunds are provided for any purchases under the applicable terms, and subscriptions can be canceled within the account, with the cancellation taking effect at the end of the current payment cycle.

API integration

Custom.MT offers public REST APIs and OpenAPI specifications, with authentication via token request headers. The permissions for full integration are specified in the Advanced and Compliance packages.

1. Create API keys in the console, and manage credentials separately for testing, production, and various systems.
2. Create and name translation templates, pre-configuring the engine, languages, terminology, editing variables before and after translation, as well as quality assessment and QA settings.
3. A text array can be submitted to the translation interface; in the newer version of this interface, it is also possible to specify tone, style, or other parameters for each individual segment.
4. Read the translation, supplier, fallback engine, quality estimation, QA, and usage fields, and implement handling logic for empty results, timeouts, and supplier errors.
5. Regularly rotate keys, restrict access levels, and monitor usage to prevent credentials from being exposed in browser code or shared files.

No official SDK released by the company was confirmed this time; developers can develop the client directly based on the API documentation.

Platforms and integration

Open source, models, and licenses

Custom.MT Console is not declared as open-source software; furthermore, there is no confirmation available regarding an official GitHub repository, an open-source license for the platform, or an official SDK.

The platform can integrate open-source machine translation models such as Meta NLLB and Helsinki NLP, and it offers services for organizing or developing such open-source models; however, the fact that it supports open-source models does not mean that its source code is available to the public.

Privacy, security, and commercial considerations

• The account privacy policy covers data related to registration, communication, usage, and payments; payment information is handled by Stripe. Personal information can be retained for the purposes of providing services throughout the duration of the account, with legal obligations and backups constituting exceptions to this rule.
• Users can request access to, correction of, or deletion of their personal information. Once an account is closed, the service provider may delete or deactivate the data stored in the active database; however, certain records may be retained for purposes related to fraud prevention, investigations, and legal obligations.
• The instructions on the MemoQ connector page state that translation data is used only during the processing phase and is not saved after the translation is completed; if third-party MT or LLL services are utilized, their handling of such data is also subject to the data protocols and privacy rules set by those providers.
• Compliance and Enterprise plans offer advanced options such as personal information masking, VPC, or on-premises deployment; some of these features are optional and do not get activated automatically upon purchasing the package.
• The terms state that the service is hosted in Germany and is governed by Czech law; it is not intended for users under 18 years of age. Cross-border teams should assess data compliance in light of their respective regions and the supplier chain.
• Users retain ownership of the content they submit, but any suggestions and feedback provided to the platform are granted to the company in accordance with the relevant terms. The copyright and commercial rights related to machine-translated output have not yet been specified; sensitive business projects should be addressed through contracts.

Suitable for users and typical use cases

• Language service providers: Create separate templates for different clients, and manage engines, terminology, tags, and costs in a unified manner.
• Software localization team: Protects interface labels, placeholders, and short codes, and obtains translations using CAT tools.
• Content and e-commerce team: Automatically feed CMS or store content into the multilingual processing workflow, followed by manual review.
• Corporate language technology team: compares different engines, integrates custom models, and uses APIs to create batch translation services.
• Organizations with high compliance requirements: Evaluate options such as tagging, VPC, or on-premises deployment to reduce the amount of sensitive data that ends up in the public cloud.

Capacity limits and usage restrictions

• Automatic metrics can only reflect specific samples and reference translations; they cannot guarantee the same quality across all fields, language pairs, and real-world content.
• The platform fees do not cover all the costs associated with translation engines; when making a budget, it is necessary to take into account the capacity of the console, any additional characters used, as well as the costs related to supplier tokens or characters.
• The more complex the workflow, the higher the actual usage resulting from prompts, terms, labels, multilingual support, and multi-stage processing.
• The availability, speed, data policies, and output rules of third-party models can change; therefore, companies should have backup engines and manual fallback procedures in place.
• The Starter version does not include explicit permissions for official API integration; to establish a technical connection, one should choose a package that lists the APIs explicitly on its page or consult with sales.

Frequently Asked Questions

Is there a permanently free version of Custom.MT?

There is no officially confirmed permanent free plan available. New users can try it out for 7 days; after that the account is suspended and an upgrade to the paid version is required to continue using it.

Does 200 euros per month cover all translation costs?

It is not included. The 200 euros for the Starter plan covers the cost of the console subscription; costs related to machine translation characters or large model tokens must be paid separately.

Can I use my own OpenAI or DeepL API key?

Yes. When a custom key is used, Custom.MT does not charge any fee for engine calls; however, the supplier will charge directly, and the processing volume still counts toward the capacity of the console plan.

Which professional translation tools are supported by Custom.MT?

Connection methods for Trados, memoQ, Smartling, XTM Cloud, Smartcat, and Shopware have been confirmed; the specific permissions depend on the package and the connector used.

Does Custom.MT provide APIs?

REST APIs and OpenAPI specifications based on token authentication are provided; the Advanced and Compliance packages include API permissions for integration.

Is Custom.MT an open-source platform?

It is not an open-source platform that has been verified. It is capable of integrating with open-source translation models, but the source code of this platform, as well as its official GitHub repository and open-source license, have not been confirmed at this time.