Yandex Translate
AITranslate plain text or HTML with the Yandex Cloud Translate API, including automatic language detection. Agents localize content without breaking the markup around it.
What This Integration Enables
Yandex Translate is the narrowest connector in this category and that is the reason to pick it. The Translate API v2 exposes exactly three methods, and this connector wraps all three. There is no document pipeline, no stored glossary to administer, no async job to poll, no quality model to configure. You give it text or HTML, it gives you back the translated version and the language it detected. Nothing else.
That makes it the right tool when translation is one step inside a workflow that is really about something else: routing a support message, publishing a product description, sending a notification to a team that reads a different language than the team that wrote it. Two details matter in practice. HTML is a first-class input, so an email body or a rich product description can be translated without its markup being treated as prose. And an inline glossary of up to 50 term pairs travels with the request, which keeps brand and product names in their original form without requiring anyone to maintain terminology in a second system. What FlowRunner adds is the part the API does not have: a place to stop. A translated string that is about to reach a customer is an automation exception waiting to happen, and agents built here know to hold it.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Answer in the language they wrote in
A message arrives in Gmail. The agent calls Detect Language on the body, and when the sender did not write in the team's working language it translates the message inbound so the support agent can read it in context. When the reply is ready, the agent translates it back into the detected language, with the glossary holding product names in place, and drafts it. The reply is a draft, not a send. The person who wrote it approves it, because the version the customer reads is not the version the person wrote and nobody should discover that after the fact.
Storefront descriptions across locales
A product description is updated in Shopify. The agent picks up the change, translates the HTML body into each locale the store sells in, and stages the result as a localized field rather than writing it live. Protected terms are supplied inline: the brand name, the collection name, the material trademarks. Anything over the 10,000 character limit for a single request is split before it is sent rather than silently truncated. The merchandiser reviews a diff and publishes.
One announcement, several reading languages
A company-wide announcement is drafted in Slack. Rather than posting it once in English and hoping, the agent translates it per regional channel and posts the localized version into each, with the original attached in a thread reply so anyone can check the source. List Supported Languages keeps the channel-to-language mapping honest as the company adds markets, instead of the mapping living in a hard-coded list that nobody updates.
Human-in-Loop Highlight
The failure this connector actually produces is structural, not semantic. When you pass HTML to Translate Text, most of the time you get well-formed HTML back. Sometimes you get output where a tag boundary moved, an attribute value was treated as prose, or an entity was rewritten, and the string still looks fine as text while rendering wrong on a page. The glossary is per-request rather than stored, so a protected term is only protected if the flow that built the request remembered to include it. Neither problem announces itself. Both reach the customer.
So the gate sits immediately before the write, not after it. The agent parses the returned markup, compares the tag structure against the source, and checks that every term in the inline glossary came back in its original form. If the structure matches and the terms held, the localized description goes live. If either check fails, the agent stops and posts to the merchandising channel: "Localized description for SKU 4417 into Turkish: source has 6 <li> elements, translation has 5, and the collection name was translated. Publish anyway, fix and re-run, or hand to a copywriter?" The choice is a person's, and it is made before a live listing changes rather than after a customer reports it. That is human-in-the-loop placed at the last reversible moment, which is the only place it is worth anything.
Agent Capabilities
3 actionsTranslation
2- Translate Text Translates one or more strings into a target language using Yandex Cloud's neural machine translation. Handles plain text or HTML, can auto-detect the source language, and returns each translation alongside the language that was detected. Supports an inline glossary of up to 50 term pairs so product names and domain vocabulary render consistently, an optional spell checker for noisy input, and custom fine-tuned models. The combined length of all strings in one request must not exceed 10,000 characters.
- Detect Language Identifies the language a text is written in and returns its Yandex language code. Accepts up to 1,000 characters. When the text is short or ambiguous, a shortlist of up to ten likely languages can be supplied to bias the detection, which materially improves accuracy for closely related languages.
Reference Data
1- List Supported Languages Retrieves every language Yandex Translate supports, each with its language code and English name. Used to populate language menus, validate a code before translating, or map a detected code back to a readable name.
Frequently Asked Questions
What can FlowRunner do with Yandex Translate?
FlowRunner agents can run Translate Text, Detect Language, and List Supported Languages in Yandex Translate.
Does connecting Yandex Translate to FlowRunner require OAuth?
No. Yandex Translate connects to FlowRunner with an API key, no OAuth flow required.
Can Yandex Translate trigger a FlowRunner workflow automatically?
Yandex Translate doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Yandex Translate
$100 in credits. No card required. Connect in minutes.