/LibreTranslate

Free and Open Source Machine Translation API. Self-hosted, offline capable and easy to setup.

Primary LanguagePythonGNU Affero General Public License v3.0AGPL-3.0

LibreTranslate

Try it online! | API Docs | Community Forum

Python versions Run tests Build and Publish Docker Image Publish package Awesome Humane Tech

Free and Open Source Machine Translation API, entirely self-hosted. Unlike other APIs, it doesn't rely on proprietary providers such as Google or Azure to perform translations. Instead, its translation engine is powered by the open source Argos Translate library.

image

Try it online! | API Docs

API Examples

Simple

Request:

const res = await fetch("https://libretranslate.com/translate", {
  method: "POST",
  body: JSON.stringify({
    q: "Hello!",
    source: "en",
    target: "es"
  }),
  headers: { "Content-Type": "application/json" }
});

console.log(await res.json());

Response:

{
    "translatedText": "¡Hola!"
}

Auto Detect Language

Request:

const res = await fetch("https://libretranslate.com/translate", {
  method: "POST",
  body: JSON.stringify({
    q: "Ciao!",
    source: "auto",
    target: "en"
  }),
  headers: { "Content-Type": "application/json" }
});

console.log(await res.json());

Response:

{
    "detectedLanguage": {
        "confidence": 83,
        "language": "it"
    },
    "translatedText": "Bye!"
}

HTML (beta)

Request:

const res = await fetch("https://libretranslate.com/translate", {
  method: "POST",
  body: JSON.stringify({
    q: '<p class="green">Hello!</p>',
    source: "en",
    target: "es",
    format: "html"
  }),
  headers: { "Content-Type": "application/json" }
});

console.log(await res.json());

Response:

{
    "translatedText": "<p class=\"green\">¡Hola!</p>"
}

Install and Run

You can run your own API server with just a few lines of setup!

Make sure you have Python installed (3.8 or higher is recommended), then simply run:

pip install libretranslate
libretranslate [args]

Then open a web browser to http://localhost:5000

On Ubuntu 20.04 you can also use the install script available at https://github.com/argosopentech/LibreTranslate-init

Run with Docker

You can also run the application with docker:

Linux/macOS

./run.sh [args]

Windows

run.bat [args]

Build and Run

See CONTIRBUTING.md for information on how to build and run the project yourself.

CUDA

You can use hardware acceleration to speed up translations on a GPU machine with CUDA 11.2 and nvidia-docker installed.

Run this version with:

docker compose -f docker-compose.cuda.yml up -d --build

Arguments

Arguments passed to the process or set via environment variables are split into two kinds.

  • Settings or runtime flags used to toggle specific runmodes or disable parts of the application. These act as toggle when added or removed.

  • Configuration parameters to set various limits and configure the application. These require a parameter to be passed to function, if removed the default parameters are used.

Settings / Flags

Argument Description Default Setting Env. name
--debug Enable debug environment Disabled LT_DEBUG
--ssl Whether to enable SSL Disabled LT_SSL
--api-keys Enable API keys database for per-client rate limits when --req-limit is reached Don't use API keys LT_API_KEYS
--require-api-key-origin Require use of an API key for programmatic access to the API, unless the request origin matches this domain No restrictions on domain origin LT_REQUIRE_API_KEY_ORIGIN
--require-api-key-secret Require use of an API key for programmatic access to the API, unless the client also sends a secret match No secrets required LT_REQUIRE_API_KEY_SECRET
--suggestions Allow user suggestions Disabled LT_SUGGESTIONS
--disable-files-translation Disable files translation File translation allowed LT_DISABLE_FILES_TRANSLATION
--disable-web-ui Disable web ui Web Ui enabled LT_DISABLE_WEB_UI
--update-models Update language models at startup Only on if no models found LT_UPDATE_MODELS
--metrics Enable the /metrics endpoint for exporting Prometheus usage metrics Disabled LT_METRICS

Configuration Parameters

Argument Description Default Parameter Env. name
--host Set host to bind the server to 127.0.0.1 LT_HOST
--port Set port to bind the server to 5000 LT_PORT
--char-limit Set character limit No limit LT_CHAR_LIMIT
--req-limit Set maximum number of requests per minute per client (outside of limits set by api keys) No limit LT_REQ_LIMIT
--req-limit-storage Storage URI to use for request limit data storage. See Flask Limiter memory:// LT_REQ_LIMIT_STORAGE
--req-time-cost Considers a time cost (in seconds) for request limiting purposes. If a request takes 10 seconds and this value is set to 5, the request cost is either 2 or the actual request cost (whichever is greater). No time cost LT_REQ_TIME_COST
--batch-limit Set maximum number of texts to translate in a batch request No limit LT_BATCH_LIMIT
--ga-id Enable Google Analytics on the API client page by providing an ID Empty (no tracking) LT_GA_ID
--frontend-language-source Set frontend default language - source auto LT_FRONTEND_LANGUAGE_SOURCE
--frontend-language-target Set frontend default language - target locale (match site's locale) LT_FRONTEND_LANGUAGE_TARGET
--frontend-timeout Set frontend translation timeout 500 LT_FRONTEND_TIMEOUT
--api-keys-db-path Use a specific path inside the container for the local database. Can be absolute or relative db/api_keys.db LT_API_KEYS_DB_PATH
--api-keys-remote Use this remote endpoint to query for valid API keys instead of using the local database Empty (use local db instead) LT_API_KEYS_REMOTE
--get-api-key-link Show a link in the UI where to direct users to get an API key Empty (no link shown on web ui) LT_GET_API_KEY_LINK
--shared-storage Shared storage URI to use for multi-process data sharing (e.g. when using gunicorn) memory:// LT_SHARED_STORAGE
--load-only Set available languages Empty (use all from argostranslate) LT_LOAD_ONLY
--threads Set number of threads 4 LT_THREADS
--metrics-auth-token Protect the /metrics endpoint by allowing only clients that have a valid Authorization Bearer token Empty (no auth required) LT_METRICS_AUTH_TOKEN
--url-prefix Add prefix to URL: example.com:5000/url-prefix/ / LT_URL_PREFIX

Notes:

  • Each argument has an equivalent environment variable that can be used instead. The env. variables overwrite the default values but have lower priority than the command arguments and are particularly useful if used with Docker. The environment variable names are the upper-snake-case of the equivalent command argument's name with a LT prefix.

  • To configure requirement for api key to use, set --req-limit to 0 and add the --api-keys flag. Requests made without a proper api key will be rejected.

  • Setting --update-models will update models regardless of whether updates are available or not.

Update

Software

If you installed with pip:

pip install -U libretranslate

If you're using docker:

docker pull libretranslate/libretranslate

Language Models

Start the program with the --update-models argument. For example: libretranslate --update-models or ./run.sh --update-models.

Alternatively you can also run the scripts/install_models.py script.

Run with WSGI and Gunicorn

pip install gunicorn
gunicorn --bind 0.0.0.0:5000 'wsgi:app'

You can pass application arguments directly to Gunicorn via:

gunicorn --bind 0.0.0.0:5000 'wsgi:app(api_keys=True)'

Kubernetes Deployment

See Medium article by JM Robles and the improved k8s.yaml by @rasos.

Helm Chart

Based on @rasos work you can now install LibreTranslate on Kubernetes using Helm.

A Helm chart is now available in the helm-chart repository where you can find more details.

You can quickly install LibreTranslate on Kubernetes using Helm with the following command:

helm repo add libretranslate https://libretranslate.github.io/helm-chart/
helm repo update
helm search repo libretranslate

helm install libretranslate libretranslate/libretranslate --namespace libretranslate --create-namespace

Manage API Keys

LibreTranslate supports per-user limit quotas, e.g. you can issue API keys to users so that they can enjoy higher requests limits per minute (if you also set --req-limit). By default all users are rate-limited based on --req-limit, but passing an optional api_key parameter to the REST endpoints allows a user to enjoy higher request limits. You can also specify different character limits that bypass the default --char-limit value on a per-key basis.

To use API keys simply start LibreTranslate with the --api-keys option. If you modified the API keys database path with the option --api-keys-db-path, you must specify the path with the same argument flag when using the ltmanage keys command.

Add New Keys

To issue a new API key with 120 requests per minute limits:

ltmanage keys add 120

To issue a new API key with 120 requests per minute and a maximum of 5,000 characters per request:

ltmanage keys add 120 --char-limit 5000

If you changed the API keys database path:

ltmanage keys --api-keys-db-path path/to/db/dbName.db add 120

Remove Keys

ltmanage keys remove <api-key>

View Keys

ltmanage keys

Prometheus Metrics

LibreTranslate has Prometheus exporter capabilities when you pass the --metrics argument at startup (disabled by default). When metrics are enabled, a /metrics endpoint is mounted on the instance:

http://localhost:5000/metrics

# HELP libretranslate_http_requests_in_flight Multiprocess metric
# TYPE libretranslate_http_requests_in_flight gauge
libretranslate_http_requests_in_flight{api_key="",endpoint="/translate",request_ip="127.0.0.1"} 0.0
# HELP libretranslate_http_request_duration_seconds Multiprocess metric
# TYPE libretranslate_http_request_duration_seconds summary
libretranslate_http_request_duration_seconds_count{api_key="",endpoint="/translate",request_ip="127.0.0.1",status="200"} 0.0
libretranslate_http_request_duration_seconds_sum{api_key="",endpoint="/translate",request_ip="127.0.0.1",status="200"} 0.0

You can then configure prometheus.yml to read the metrics:

scrape_configs:
  - job_name: "libretranslate"

    # Needed only if you use --metrics-auth-token
    #authorization:
      #credentials: "mytoken"

    static_configs:
      - targets: ["localhost:5000"]

To secure the /metrics endpoint you can also use --metrics-auth-token mytoken.

If you use Gunicorn, make sure to create a directory for storing multiprocess data metrics and set PROMETHEUS_MULTIPROC_DIR:

mkdir -p /tmp/prometheus_data
rm /tmp/prometheus_data/*
export PROMETHEUS_MULTIPROC_DIR=/tmp/prometheus_data
gunicorn -c scripts/gunicorn_conf.py --bind 0.0.0.0:5000 'wsgi:app(metrics=True)'

Language Bindings

You can use the LibreTranslate API using the following bindings:

Discourse Plugin

You can use the official discourse translator plugin to translate Discourse topics with LibreTranslate. To install it simply modify /var/discourse/containers/app.yml:

## Plugins go here
## see https://meta.discourse.org/t/19157 for details
hooks:
  after_code:
    - exec:
        cd: $home/plugins
        cmd:
          - git clone https://github.com/discourse/docker_manager.git
          - git clone https://github.com/discourse/discourse-translator
    ...

Then issue ./launcher rebuild app. From the Discourse's admin panel then select "LibreTranslate" as a translation provider and set the relevant endpoint configurations.

See it in action on this page.

Mobile Apps

Web browser

Mirrors

This is a list of public LibreTranslate instances, some require an API key. If you want to add a new URL, please open a pull request.

URL API Key Required Links
libretranslate.com ✔️ [ Get API Key ] [ Service Status ]
translate.terraprint.co -
trans.zillyhuhn.com -
libretranslate.eownerdead.dedyn.io -

TOR/i2p Mirrors

URL
lt.vernccvbvyi5qhfzyqengccj7lkove6bjot2xhh5kajhwvidqafczrad.onion
lt.vern.i2p

Adding New Language Models

You have two options to create new language models:

Most of the training data is from Opus, which is an open source parallel corpus. Check also NLLU

Localization

The LibreTranslate Web UI is available in all the languages for which LibreTranslate can translate to. It can also (roughly) translate itself! Some languages might not appear in the UI since they haven't been reviewed by a human yet. You can enable all languages by turning on --debug mode.

To help improve or review the UI translations:

{
 "name": "<Language>",
 "reviewed": true <-- Change this from false to true
}

UI Languages

Language Reviewed Weblate Link
Arabic Edit
Azerbaijani Edit
Chinese Edit
Chinese (Traditional) Edit
Czech ✔️ Edit
Danish Edit
Dutch Edit
English ✔️ Edit
Esperanto ✔️ Edit
Finnish Edit
French ✔️ Edit
German ✔️ Edit
Greek Edit
Hebrew Edit
Hindi Edit
Hungarian Edit
Indonesian Edit
Irish Edit
Italian ✔️ Edit
Japanese Edit
Kabyle Edit
Korean ✔️ Edit
Occitan Edit
Persian Edit
Polish Edit
Portuguese ✔️ Edit
Russian ✔️ Edit
Slovak Edit
Spanish ✔️ Edit
Swedish Edit
Turkish Edit
Ukranian ✔️ Edit
Vietnamese Edit

Roadmap

Help us by opening a pull request!

Any other idea is welcome also.

FAQ

Can I use your API server at libretranslate.com for my application in production?

In short, yes, but only if you buy an API key. You can always run LibreTranslate for free on your own server of course.

Some translations on libretranslate.com are different than the self-hosted ones. Why?

By default language models are loaded from the argos-index. Sometimes we deploy models on libretranslate.com that haven't been added to the argos-index yet, such as those converted from OPUS (thread)

Where are the language models saved?

In $HOME/.local/share/argos-translate/packages. On Windows that's C:\Users\youruser\.local\share\argos-translate\packages.

Can I use LibreTranslate behind a reverse proxy, like Apache2 or Caddy?

Yes, here are config examples for Apache2 and Caddy that redirect a subdomain (with HTTPS certificate) to LibreTranslate running on a docker at localhost.

sudo docker run -ti --rm -p 127.0.0.1:5000:5000 libretranslate/libretranslate

You can remove 127.0.0.1 on the above command if you want to be able to access it from domain.tld:5000, in addition to subdomain.domain.tld (this can be helpful to determine if there is an issue with Apache2 or the docker container).

Add --restart unless-stopped if you want this docker to start on boot, unless manually stopped.

Apache config

Replace [YOUR_DOMAIN] with your full domain; for example, translate.domain.tld or libretranslate.domain.tld.

Remove # on the ErrorLog and CustomLog lines to log requests.

#Libretranslate

#Redirect http to https
<VirtualHost *:80>
    ServerName http://[YOUR_DOMAIN]
    Redirect / https://[YOUR_DOMAIN]
    # ErrorLog ${APACHE_LOG_DIR}/error.log
    # CustomLog ${APACHE_LOG_DIR}/tr-access.log combined
 </VirtualHost>

#https
<VirtualHost *:443>
    ServerName https://[YOUR_DOMAIN]

    ProxyPass / http://127.0.0.1:5000/
    ProxyPassReverse / http://127.0.0.1:5000/
    ProxyPreserveHost On

    SSLEngine on
    SSLCertificateFile /etc/letsencrypt/live/[YOUR_DOMAIN]/fullchain.pem
    SSLCertificateKeyFile /etc/letsencrypt/live/[YOUR_DOMAIN]/privkey.pem
    SSLCertificateChainFile /etc/letsencrypt/live/[YOUR_DOMAIN]/fullchain.pem

    # ErrorLog ${APACHE_LOG_DIR}/tr-error.log
    # CustomLog ${APACHE_LOG_DIR}/tr-access.log combined
</VirtualHost>

Add this to an existing site config, or a new file in /etc/apache2/sites-available/new-site.conf and run sudo a2ensite new-site.conf.

To get a HTTPS subdomain certificate, install certbot (snap), run sudo certbot certonly --manual --preferred-challenges dns and enter your information (with subdomain.domain.tld as the domain). Add a DNS TXT record with your domain registrar when asked. This will save your certificate and key to /etc/letsencrypt/live/{subdomain.domain.tld}/. Alternatively, comment the SSL lines out if you don't want to use HTTPS.

Caddy config

Replace [YOUR_DOMAIN] with your full domain; for example, translate.domain.tld or libretranslate.domain.tld.

#Libretranslate
[YOUR_DOMAIN] {
  reverse_proxy localhost:5000
}

Add this to an existing Caddyfile or save it as Caddyfile in any directory and run sudo caddy reload in that same directory.

NGINX config

Replace [YOUR_DOMAIN] with your full domain; for example, translate.domain.tld or libretranslate.domain.tld.

Remove # on the access_log and error_log lines to disable logging.

server {
  listen 80;
  server_name [YOUR_DOMAIN];
  return 301 https://$server_name$request_uri;
}

server {
  listen 443 http2 ssl;
  server_name [YOUR_DOMAIN];

  #access_log off;
  #error_log off;

  # SSL Section
  ssl_certificate /etc/letsencrypt/live/[YOUR_DOMAIN]/fullchain.pem;
  ssl_certificate_key /etc/letsencrypt/live/[YOUR_DOMAIN]/privkey.pem;

  ssl_protocols TLSv1.2 TLSv1.3;

  # Using the recommended cipher suite from: https://wiki.mozilla.org/Security/Server_Side_TLS
  ssl_ciphers 'ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-CHACHA20-POLY1305:ECDHE-RSA-CHACHA20-POLY1305:DHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES256-GCM-SHA384';

  ssl_session_timeout 10m;
  ssl_session_cache shared:MozSSL:10m;  # about 40000 sessions
  ssl_session_tickets off;

  # Specifies a curve for ECDHE ciphers.
  ssl_ecdh_curve prime256v1;
  # Server should determine the ciphers, not the client
  ssl_prefer_server_ciphers on;


  # Header section
  add_header Strict-Transport-Security  "max-age=31536000; includeSubDomains; preload" always;
  add_header Referrer-Policy            "strict-origin" always;

  add_header X-Frame-Options            "SAMEORIGIN"    always;
  add_header X-XSS-Protection           "1; mode=block" always;
  add_header X-Content-Type-Options     "nosniff"       always;
  add_header X-Download-Options         "noopen"        always;
  add_header X-Robots-Tag               "none"          always;

  add_header Feature-Policy             "microphone 'none'; camera 'none'; geolocation 'none';"  always;
  # Newer header but not everywhere supported
  add_header Permissions-Policy         "microphone=(), camera=(), geolocation=()" always;

  # Remove X-Powered-By, which is an information leak
  fastcgi_hide_header X-Powered-By;

  # Do not send nginx server header
  server_tokens off;

  # GZIP Section
  gzip on;
  gzip_disable "msie6";

  gzip_vary on;
  gzip_proxied any;
  gzip_comp_level 6;
  gzip_buffers 16 8k;
  gzip_http_version 1.1;
  gzip_min_length 256;
  gzip_types text/xml text/javascript font/ttf font/eot font/otf application/x-javascript application/atom+xml application/javascript application/json application/manifest+json application/rss+xml application/x-web-app-manifest+json application/xhtml+xml application/xml image/svg+xml image/x-icon text/css text/plain;

  location / {
      proxy_pass http://127.0.0.1:5000/;
      proxy_set_header Host $http_host;
      proxy_set_header X-Real-IP $remote_addr;
      proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
      proxy_set_header X-Forwarded-Proto $scheme;
      client_max_body_size 0;
  }
}

Add this to an existing NGINX config or save it as libretranslate in the /etc/nginx/site-enabled directory and run sudo nginx -s reload.

Can I do batch translations?

Yes, pass an array of strings instead of a string to the q field:

const res = await fetch("https://libretranslate.com/translate", {
  method: "POST",
  body: JSON.stringify({
    q: ["Hello", "world"],
    source: "en",
    target: "es"
  }),
  headers: { "Content-Type": "application/json" }
});

console.log(await res.json());
// {
//     "translatedText": [
//         "Hola",
//         "mundo"
//     ]
// }

Contributing

We welcome contributions! Here's some ideas:

  • Train a new language model using Locomotive. For example, we want to train improved neural networks for German and many other languages.
  • Can you beat the performance of our language models? Train a new one and let's compare it. To submit your model make a post on the community forum with a link to download your .argosmodel file and some sample text that your model has translated.
  • Pick an issue to work on.

Credits

This work is largely possible thanks to Argos Translate, which powers the translation engine.

License

GNU Affero General Public License v3

Trademark

See Trademark Guidelines